Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Brouwer fixed-point theorem wikipedia , lookup
Sheaf (mathematics) wikipedia , lookup
Homotopy type theory wikipedia , lookup
Symmetric space wikipedia , lookup
Grothendieck topology wikipedia , lookup
Fundamental group wikipedia , lookup
Covering space wikipedia , lookup
699 Documenta Math. Thom Spectra that Are Symmetric Spectra Christian Schlichtkrull Received: November 7, 2007 Revised: December 18, 2009 Communicated by Stefan Schwede Abstract. We analyze the functorial and multiplicative properties of the Thom spectrum functor in the setting of symmetric spectra and we establish the relevant homotopy invariance. 2000 Mathematics Subject Classification: 55P43 Keywords and Phrases: Thom spectra, symmetric spectra. 1. Introduction The purpose of this paper is to develop the theory of Thom spectra in the setting of symmetric spectra. In particular, we establish the relevant homotopy invariance and we investigate the multiplicative properties. Classically, given a sequence of spaces X0 → X1 → X2 → . . . , equipped with a compatible sequence of maps fn : Xn → BO(n), the Thom spectrum T (f ) is defined by pulling the universal bundles V (n) over BO(n) back via the fn ’s and letting T (f )n = f ∗ V (n)/Xn . Here the bar denotes fibre-wise one-point compactification. More generally, one may consider compatible families of maps Xn → BF (n), where F (n) is the topological monoid of base point preserving self-homotopy equivalences of S n , and similarly define a Thom spectrum by pulling back the canonical S n -(quasi)fibration over BF (n). Composing with the canonical maps BO(n) → BF (n), one sees that the latter construction generalizes the former. This generalization was suggested by Mahowald [24], [25], and has been investigated in detail by Lewis in [20]. Documenta Mathematica 14 (2009) 699–748 700 Christian Schlichtkrull 1.1. Symmetric Thom spectra via I-spaces. In order to translate the definition of Thom spectra into the setting of symmetric spectra, we shall modify the construction by considering certain diagrams of spaces. Let I be the category whose objects are the finite sets n = {1, . . . , n}, together with the empty set 0, and whose morphisms are the injective maps. The concatenation m ⊔ n in I is defined by letting m correspond to the first m and n to the last n elements of {1, . . . , m + n}. This makes I a symmetric monoidal category with symmetric structure given by the (m, n)-shuffles τm,n : m ⊔ n → n ⊔ m. We define an I-space to be a functor from I to the category U of spaces and write IU for the category of such functors. The correspondence n → BF (n) defines an I-space and we show that if X → BF is a map of I-spaces, then the Thom spectrum T (f ) defined as above is a symmetric spectrum. The main advantage of the category SpΣ of symmetric spectra to ordinary spectra is that it has a symmetric monoidal smash product. Similarly, the category IU/BF of I-spaces over BF inherits a symmetric monoidal structure from I and the Thom spectrum functor is compatible with these structures in the following sense. Theorem 1.1. The symmetric Thom spectrum functor T : IU/BF → SpΣ is strong symmetric monoidal. That T is strong symmetric monoidal means of course that there is a natural isomorphism of symmetric spectra T (f ) ∧ T (g) ∼ = T (f ⊠ g), where f ⊠ g denotes the monoidal product in IU/BF . In particular, T takes monoids in IU/BF to symmetric ring spectra. A similar construction can be carried out in the setting of orthogonal spectra and the idea of realizing Thom spectra as “structured ring spectra” by such a diagrammatic approach goes back to [31]. 1.2. Lifting space level data to I-spaces. Let N be the ordered set of non-negative integers, thought of as a subcategory of I via the canonical subset inclusions. Another starting point for the construction of Thom spectra is to consider maps X → BFN , where BFN denotes the colimit of the I-space BF restricted to N . Given such a map, one may choose a suitable filtration of X so as to get a map of N -spaces X(n) → BF (n) and the definition of the Thom spectrum T (f ) then proceeds as above. This is the point of view taken by Lewis [20]. The space BFN has an action of the linear isometries operad L, and Lewis proves that if f is a map of C-spaces where C is an operad that is augmented over L, then the Thom spectrum T (f ) inherits an action of C. In the setting of symmetric spectra the problem is how to lift space level data to objects in IU/BF . We think of I as some kind of algebraic structure acting on BF , and in order to pull such an action back via a space level map we should ideally map into the quotient space BFI , that is, into the colimit over I. The problem with this approach is that the homotopy type of BFI differs from that of BFN . For this reason we shall instead work with the homotopy colimit BFhI which does have the correct homotopy type. We prove in Section 4 that the homotopy colimit functor has a right adjoint U : U/BFhI → IU/BF such Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 701 that this pair of adjoint functors defines a Quillen equivalence hocolim : IU/BF o I / U/BF : U. hI Here the model structure on IU is the one established by Sagave-Schlichtkrull [33]. The weak equivalences in this model structure are called I-equivalences and are the maps that induce weak homotopy equivalences on the associated homotopy colimits; see Section 4.1 for details. It follows from the theorem that the homotopy theory associated to IU/BF is equivalent to that of U/BFhI . As is often the case for functors that are right adjoints, U is only homotopically well-behaved when applied to fibrant objects. We shall usually remedy this by composing with a suitable fibrant replacement functor on U/BFhI and we write U ′ for the composite functor so defined. Composing the right adjoint U with the symmetric Thom spectrum functor from Theorem 1.1 we get a Thom spectrum functor on U/BFhI . However, even when restricted to fibrant objects this functor does not have all the properties one may expect from a Thom spectrum functor. Notably, one of the important properties of the Lewis-May Thom spectrum functor on U/BFN is that it preserves colimits whereas the symmetric Thom spectrum obtained by composing with U does not have this property. For this reason we shall introduce another procedure for lifting space level data to I-spaces in the form of a functor R : U/BFhI → IU/BF and we shall use this functor to associate Thom spectra to objects in U/BFhI . The first statement in the following theorem ensures that the functor so defined produces Thom spectra with the correct homotopy type. ∼ Theorem 1.2. There is a natural level equivalence R − → U ′ over BF and the symmetric Thom spectrum functor defined by the composition R T → IU/BF − → SpΣ T : U/BFhI − preserves colimits. As indicated in the theorem we shall use the notation T both for the symmetric Thom spectrum on IU/BF and for its composition with R; the context will always make the meaning clear. In Section 4.4 we show that in a precise sense our Thom spectrum functor becomes equivalent to that of Lewis-May when composing with the forgetful functor from symmetric spectra to spectra. We also have the following analogue of Lewis’ result imposing L-actions on Thom spectra. In our setting the relevant operad is the Barrat-Eccles operad E, see [2] and [28], Remarks 6.5. We recall that E is an E∞ operad and that a space with an E-action is automatically an associative monoid. Theorem 1.3. The operad E acts on BFhI and if f : X → BFhI is a map of C-spaces where C is an operad that is augmented over E, then T (f ) inherits an action of C. Documenta Mathematica 14 (2009) 699–748 702 Christian Schlichtkrull We often find that the enriched functoriality obtained by working with homotopy colimits over I instead of colimits over N is very useful. For example, one may represent complexification followed by realification as maps of E-spaces BOhI → BUhI → BOhI , such that the composite E∞ map represents multiplication by 2. The procedure for lifting space level data described above works quite generally for diagram categories. Implemented in the framework of orthogonal spectra, it gives an answer to the problem left open in [32], Chapter 23, on how to construct orthogonal Thom spectra from space level data; we spell out the details of this in Section 8.5. We also remark that one can define an I-space BGL1 (A) for any symmetric ring spectrum A, and that an analogous lifting procedure allows one to associate A-module Thom spectra to maps X → BGL1 (A)hI . We hope to return to this in a future paper. 1.3. Homotopy invariance. Ideally, one would like the symmetric Thom spectrum functor to take I-equivalences of I-spaces over BF to stable equivalences of symmetric spectra. However, due to the fact that quasifibrations are not in general preserved under pullbacks this is not true without further assumptions on the objects in IU/BF . We say that an object (X, f ) (that is, a map f : X → BF ) is T -good if T (f ) has the same homotopy type as the Thom spectrum associated to a fibrant replacement of f ; see Definition 5.1 for details. Theorem 1.4. If (X, f ) → (Y, g) is an I-equivalence of T -good I-spaces over BF , then the induced map T (f ) → T (g) is a stable equivalence of symmetric spectra. Here stable equivalence refers to the stable model structure on SpΣ defined in [16] and [27]. It is a subtle property of this model structure that a stable equivalence needs not induce an isomorphism of stable homotopy groups. However, if X and Y are convergent (see Section 2), then the associated Thom spectra are also convergent, and in this case a stable equivalence is indeed a π∗ isomorphism in the usual sense. The T -goodness requirement in the theorem is not a real restriction since in general any object in U/BF can (and should) be replaced by one that is T -good. The functor R takes values in the subcategory of convergent T -good objects and takes weak homotopy equivalences to I-equivalences (in fact to level-wise equivalences). It follows that the Thom spectrum functor in Theorem 1.2 is a homotopy functor; see Corollary 4.13. Example 1.5. Theorem 1.4 also has interesting consequences for Thom spectra that are not convergent. As an example, consider the Thom spectrum M O(1)∧∞ that represents the bordism theory of manifolds whose stable normal bundle splits as a sum of line bundles, see [1], [9]. This is the symmetric Thom spectrum associated to the map of I-spaces X(n) → BF (n), where X(n) = BO(1)n . It is proved in [34] that XhI is homotopy equivalent to Q(RP ∞ ), hence it follows that M O(1)∧∞ is stably equivalent as a symmetric Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 703 ring spectrum to the Thom spectrum associated to the map of infinite loop spaces Q(RP ∞ ) → BFhI . In general any I-space X is I-equivalent to the constant I-space XhI and consequently any symmetric Thom spectrum is stably equivalent to one arising from a space-level map. However, the added flexibility obtained by working in IU is often very convenient. Notably, it is proved in [33] that any E∞ monoid in IU is equivalent to a strictly commutative monoid; something which is wellknown not to be the case in U. 1.4. Applications to the Thom isomorphism. As an application of the techniques developed in this paper we present a strictly multiplicative version of the Thom isomorphism. A map f : X → BFhI gives rise to a morphism in U/BFhI , ∆ : (X, f ) → (X × X, f ◦ π2 ), where ∆ is the diagonal inclusion and π2 denotes the projection onto the second factor of X × X. The Thom spectrum T (f ◦ π2 ) is isomorphic to X+ ∧ T (f ), and the Thom diagonal ∆ : T (f ) → X+ ∧ T (f ) is the map of Thom spectra induced by ∆. In Section 7.1 we define a canonical orientation T (f ) → H, where H denotes (a convenient model of) the EilenbergMac Lane spectrum HZ/2. Using this we get a map of symmetric spectra (1.6) ∆∧H T (f ) ∧ H −−−→ X+ ∧ T (f ) ∧ H → X+ ∧ H ∧ H → X+ ∧ H, where the last map is induced by the multiplication in H. The spectrum level version of the Z/2-Thom isomorphism theorem is the statement that this is a stable equivalence, see [26]. If f is oriented in the sense that it lifts to a map f : X → BSFhI , then we define a canonical integral orientation T (f ) → HZ and the spectrum level version of the integral Thom isomorphism theorem is the statement that the induced map (1.7) T (f ) ∧ HZ → X+ ∧ HZ is a stable equivalence. In our framework these results lift to “structured ring spectra” in the sense of the following theorem. Here C again denotes an operad that is augmented over E. Theorem 1.8. If f : X → BFhI (respectively f : X → BSFhI ) is a map of C-spaces, then the spectrum level Thom equivalence (1.6) (respectively (1.7)) is a C-map. For example, one may represent the complex cobordism spectrum M U as the Thom spectrum associated to the E-map BUhI → BSFhI and the Thom equivalence (1.7) is then an equivalence of E∞ symmetric ring spectra. This should be compared with the H∞ version in [20]. Documenta Mathematica 14 (2009) 699–748 704 Christian Schlichtkrull 1.5. Diagram Thom spectra and symmetrization. The definition of the symmetric Thom spectrum functor shows that the category I is closely related to the category of symmetric spectra. However, many of the Thom spectra that occur in the applications do not naturally arise from a map of I-spaces but rather from a map of D-spaces for some monoidal category D equipped with a monoidal functor D → I. We formalize this in Section 8 where we introduce the notion of a D-spectrum associated to such a monoidal functor. For example, the complex cobordism spectrum M U associated to the unitary groups U (n) and the Thom spectrum M B associated to the braid groups B(n) can be realized as diagram ring spectra in this way. It is often convenient to replace the D-Thom spectrum associated to a map of D-spaces f : X → BF by a symmetric spectrum, and our preferred way of doing this is to first transform f to a map of I-spaces and then evaluate the symmetric Thom spectrum functor on this transformed map. In this way we end up with a symmetric spectrum to which we can exploit the structural relationship to the category of I-spaces. We shall discuss various ways of carrying out this “symmetrization” process and in particular we shall see how to realize the Thom spectra M U and M B as (in the case of M U commutative) symmetric ring spectra. 1.6. Organization of the paper. We begin by recalling the basic facts about Thom spaces and Thom spectra in Section 2, and in Section 3 we introduce the symmetric Thom spectrum functor and show that it is strong symmetric monoidal. The I-space lifting functor R is introduced in Section 4, where we prove Theorem 1.2 in a more precise form; this is the content of Proposition 4.10 and Corollary 4.13. Here we also compare the Lewis-May Thom spectrum functor to our construction. We prove the homotopy invariance result Theorem 1.4 in Section 5, and in Section 6 we analyze to what extent the constructions introduced in the previous sections are preserved under operad actions. In particular, we prove Theorem 1.3 in a more precise form; this is the content of Corollary 6.9. The Thom isomorphism theorem is proved in Section 7 and in Section 8 we discuss how to symmetrize other types of diagram Thom spectra and how the analogue of the lifting functor R works in the context of orthogonal spectra. Finally, we have included some background material on homotopy colimits in Appendix A. 1.7. Notation and conventions. We shall work in the categories U and T of unbased and based compactly generated weak Hausdorff spaces. By a cofibration we understand a map having the homotopy extension property, see [39]. A based space is well-based if the inclusion of the base point is a cofibration. In this paper S n always denotes the one-point compactification of Rn . By a spectrum E we understand a sequence {En : n ≥ 0} of based spaces together with a sequence of based structure maps S 1 ∧ En → En+1 . A map of spectra f : E → F is a sequence of based maps fn : En → Fn that commute with the structure maps and we write Sp for the category of spectra so defined. A spectrum is connective if πn (E) = 0 for n < 0 and convergent if there is Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 705 an unbounded, non-decreasing sequence of integers {λn : n ≥ 0} such that the adjoint structure maps En → ΩEn+1 are (λn + n)-connected for all n. 2. Preliminaries on Thom spaces and Thom spectra In this section we recall the basic facts about Thom spaces and Thom spectra that we shall need. The main reference for this material is Lewis’ account in [20], Section IX. Here we emphasize the details relevant for the construction of symmetric Thom spectra in Section 3. We begin by recalling the two-sided simplicial bar construction and some of its properties, referring to [30] for more details. Given a topological monoid G, a right G-space Y , and a left G-space X, this is the simplicial space B• (Y, G, X) with k-simplices Y × Gk × X and simplicial operators for i = 0 (yg1 , g2 . . . , gk , x), di (y, g1 , . . . , gk , x) = (y, g1 , . . . , gi gi+1 , . . . , gk , x), for 0 < i < k (y, g1 , . . . , gk x), for i = k, and si (y, g1 , . . . , gk , x) = (y, . . . , gi−1 , 1, gi , . . . , x), for 0 ≤ i ≤ k. We write B(Y, G, X) for the topological realization. In the case where X and Y equal the one-point space ∗, this is the usual simplicial construction of the classifying space BG. The projection of X onto ∗ induces a map p : B(Y, G, X) → B(Y, G, ∗) whose fibres are homeomorphic to X. Furthermore, if X has a G-invariant basepoint, then the inclusion of the base point defines a section s : B(Y, G, ∗) → B(Y, G, X). Recall that a topological monoid is grouplike if the set of components with the induced monoid structure is a group. Proposition 2.1 ([19],[30]). If G is a well-based grouplike monoid, then the projection p is a quasifibration, and if X has a G-invariant base point such that X is (non-equivariantly) well-based, then the section s is a cofibration. In general we say that a sectioned quasifibration is well-based if the section is a cofibration. Let F (n) be the topological monoid of base point preserving homotopy equivalences of S n , where we recall that the latter denotes the onepoint compactification of Rn . It follows from [19], Theorem 2.1, that this is a well-based monoid and we let V (n) = B(∗, F (n), S n ). Then BF (n) is a classifying space for sectioned fibrations with fibre equivalent to S n and the projection pn : V (n) → BF (n) is a well-based quasifibration. Given a map f : X → BF (n), let pX : f ∗ V (n) → X be the pull-back of V (n) along f , and notice that the section s gives rise to a section sX : X → f ∗ V (n). The associated Thom space is the quotient space T (f ) = f ∗ V (n)/sX (X). Documenta Mathematica 14 (2009) 699–748 706 Christian Schlichtkrull This construction is clearly functorial on the category U/BF (n) of spaces over BF (n). We often use the notation (X, f ) for an object f : X → BF (n) in this category. In order for the Thom space functor to be homotopically wellbehaved we would like pX to be a quasifibration and sX to be a cofibration, but unfortunately this is not true in general. This is the main technical difference compared to working with sectioned fibrations. For our purpose it will not do to replace the quasifibration pn by an equivalent fibration since we then loose the strict multiplicative properties of the bar construction required for the definition of strict multiplicative structures on Thom spectra. We say that f classifies a well-based quasifibration if pX is a quasi-fibration and sX is a cofibration. The following well-known results are included here for completeness. Lemma 2.2 ([20]). Given a well-based sectioned quasifibration p : V → B and a Hurewicz fibration f : X → B, the pullback pX : f ∗ V → X is again a wellbased quasifibration. Proof. Since f is a fibration the pullback diagram defining f ∗ V is homotopy cartesian, hence f ∗ V → X is a quasifibration. In order to see that the section sX is a cofibration, notice that it is the pullback of the section of p along the Hurewicz fibration f ∗ V → V . The result then follows from Theorem 12 of [40] which states that the pullback of a cofibration along a Hurewicz fibration is again a cofibration. Let Top(n) be the topological group of base point preserving homeomorphisms of S n . The next result is the main reason why the objects in U/BF (n) that factor through BTop(n) are easier to handle than general objects. Lemma 2.3. If f : X → BF (n) factors through BTop(n), then f classifies a well-based Hurewicz fibration, hence a well-based quasifibration. Proof. Let W (n) = B(∗, Top(n), S n ). The projection W (n) → BTop(n) is a fibre bundle by [30], Corollary 8.4, and in particular a Hurewicz fibration. Suppose that f factors through a map g : X → BTop(n). Then f ∗ V (n) is homeomorphic to g ∗ W (n) and thus pX is a Hurewicz fibration. We must prove that the section is a cofibration. Let us use the Strøm model structure [41] on U to get a factorization g = g2 g1 where g1 is a cofibration and g2 is a Hurewicz fibration. From this we get a factorization of the pullback diagram defining g ∗ W (n), / W (n) / g ∗ W (n) g ∗ W (n) 2 pX X pY g1 /Y g2 / BTop(n), and it follows from Lemma 2.2 that the section sY of pY is a cofibration. Since pY is a Hurewicz fibration it follows by the same argument that the induced map g ∗ W (n) → g2∗ W (n) is also a cofibration. It is clear that the composition X → g ∗ W (n) → g2∗ W (n) is a cofibration and the conclusion thus follows from Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 707 Lemma 5 of [41], which states that if h = i ◦ j is a composition of maps in which h and i are both cofibrations, then j is a cofibration as well. This lemma applies in particular if f factors through BO(n). In order to get around the difficulty that the Thom space functor is not a homotopy functor on the whole category U/BF (n) we follow Lewis [20], Section IX, and define a functor (2.4) Γ : U/BF (n) → U/BF (n), (X, f ) 7→ (Γf (X), Γ(f )) by replacing a map by a (Hurewicz) fibration in the usual way, Γf (X) = {(x, ω) ∈ X × BF (n)I : f (x) = ω(0)}, Γ(f )(x, ω) = ω(1). We sometimes write Γ(X) instead of Γf (X) when the map f is clear from the context. The natural inclusion X → Γf (X), whose second coordinate is the constant path at f (x), defines a natural equivalence from the identity functor on U/BF (n) to Γ. It follows from Lemma 2.2 that the composition of the Thom space functor T with Γ is a homotopy functor. We think of T (Γ(f )) as representing the correct homotopy type of the Thom space and say that f is T -good if the natural map T (f ) → T (Γ(f )) is a weak homotopy equivalence. In particular, f is T -good if it classifies a well-based quasifibration. It follows from the above discussion that the restriction of T to the subcategory of T -good objects is a homotopy functor. Remark 2.5. The fibrant replacement functors used in [20] and [30] are defined using Moore paths instead of paths defined on the unit interval I. The use of Moore paths is less convenient for our purposes since we shall use Γ in combination with more general homotopy pullback constructions. The following basic lemma is needed in order to establish the connectivity and convergence properties of the Thom spectrum functor. It may for example be deduced from the dual Blakers-Massey Theorem in [14]. Lemma 2.6. Let V1 −−−−→ p1 y β V2 p2 y B1 −−−−→ B2 be a pullback diagram of well-based quasifibrations p1 and p2 . Suppose that p1 and p2 are n-connected with n > 1 and that β is k-connected. Then the quotient spaces V1 /B1 and V2 /B2 are well-based and (n − 1)-connected, and the induced map V1 /B1 → V2 /B2 is (k + n)-connected. We now turn to Thom spectra. Let N be as in Section 1.2, and write N U for the category of N -spaces, that is, functors X : N → U. Consider the N -space BF defined by the sequence of cofibrations i i i 2 1 0 ... BF (2) → BF (1) → BF (0) → Documenta Mathematica 14 (2009) 699–748 708 Christian Schlichtkrull obtained by applying B to the monoid homomorphisms F (n) → F (n + 1) that take an element u to 1S 1∧u, the smash product with the identity on S 1 . Notice, that there are pullback diagrams ¯ V (n) −−−−→ V (n + 1) S1∧ (2.7) y y BF (n) i −−−n−→ BF (n + 1), ¯ − denotes fibre-wise smash product with S 1 . Indeed, there clearly where S 1 ∧ is such a pullback diagram of the underlying simplicial spaces, and topological realization preserves pullback diagrams. We let N U/BF be the category of N -spaces over BF . Thus, an object is a sequence of maps fn : X(n) → BF (n) that are compatible with the structure maps. Again we may specify the domain by writing the objects in the form (X, f ). Definition 2.8. The Thom spectrum functor T : N U/BF → Sp is defined by applying the Thom space construction level-wise, T (f )n = T (fn ), with structure maps given by S 1 ∧ T (fn ) ∼ = T (in ◦ fn ) → T (fn+1 ). A morphism in N U/BF induces a map of Thom spectra in the obvious way. As for the Thom space functor, the Thom spectrum functor is not homotopically well-behaved on the whole category N U/BF . We define a functor Γ : N U/BF → N U/BF by applying the functor Γ level-wise, and we say that an object (X, f ) is T -good if the induced map T (f ) → T (Γ(f )) is a stable equivalence. We say that f is level-wise T -good if the induced map is a levelwise equivalence. The following proposition is an immediate consequence of Lemma 2.2 and Lemma 2.6. Proposition 2.9. If f : X → BF is T -good, then T (f ) is connective. An N -space X is said to be convergent if there exists an unbounded, nondecreasing sequence of integers {λn : n ≥ 0} such that X(n) → X(n + 1) is λn -connected for each n. Proposition 2.10. If f : X → BF is level-wise T -good and X is convergent, then T (f ) is also convergent. Proof. We may assume that f is a level-wise fibration, hence classifies a wellbased quasifibration at each level. If X(n) → X(n + 1) is λn -connected, it follows from Lemma 2.6 that the structure map S 1 ∧ T (fn ) → T (fn+1 ) is (λn + n)-connected. The convergence of X thus implies that of T (f ). Given an N -space X, write XhN for its homotopy colimit. This is homotopy equivalent to the usual telescope construction on X. We say that a morphism (X, f ) → (Y, g) in N U/BF is an N -equivalence if the induced map of homotopy Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 709 colimits XhN → YhN is a weak homotopy equivalence. The following theorem can be deduced from [20], Proposition 4.9. We shall indicate a more direct proof in Section 5.1. Theorem 2.11. If (X, f ) → (Y, g) is an N -equivalence of T -good N -spaces over BF , then the induced map T (f ) → T (g) is a stable equivalence. In particular, it follows that T ◦ Γ takes N -equivalences to stable equivalences. 3. Symmetric Thom spectra We begin by recalling the definition of a (topological) symmetric spectrum. The basic references are the papers [16] and [27] that deal respectively with the simplicial and the topological version of the theory. See also [35]. 3.1. Symmetric spectra. By definition a symmetric spectrum X is a spectrum in which each of the spaces X(n) is equipped with a base point preserving left Σn -action, such that the iterated structure maps σ n : S m ∧ X(n) → X(m + n) are Σm ×Σn -equivariant. A map of symmetric spectra f : X → Y is a sequence of Σn -equivariant based maps X(n) → Y (n) that strictly commute with the structure maps. We write SpΣ for the category of symmetric spectra. Following [27] we shall view symmetric spectra as diagram spectra, and for this reason we introduce some notation which will be convenient for our purposes. Let the category I be as in Section 1.1. Given a morphism α : m → n in I, let n − α denote the complement of α(m) in n and let S n−α be the one-point compactification of Rn−α . Consider then the topological category IS that has the same objects as I, but whose morphism spaces are defined by _ S n−α . IS (m, n) = α∈I(m,n) We view IS as a category enriched in the category of based spaces T . Writing the morphisms in the form (x, α) for x ∈ S n−α , the composition is defined by IS (m, n) ∧ IS (l, m) → IS (l, n), (x, α) ∧ (y, β) 7→ (x ∧ α∗ y, αβ), where x ∧ α∗ y is defined by the canonical homeomorphism S n−α ∧ S m−β ∼ = S n−αβ , x ∧ y 7→ x ∧ α∗ y, obtained by reindexing the coordinates of S m−β via α. This choice of notation has the advantage of making some of our constructions self-explanatory. By a functor between categories enriched in T we understand a functor such that the maps of morphism spaces are based and continuous. Thus, if X : IS → T is a functor in this sense, then we have for each morphism α : m → n in I a based continuous map α∗ : S n−α ∧ X(m) → X(n). One easily checks that the maps S 1 ∧ X(n) → X(n + 1) induced by the morphisms n → 1 ⊔ n give X the structure of a symmetric spectrum and that the Documenta Mathematica 14 (2009) 699–748 710 Christian Schlichtkrull category of (based continuous) functors IS → T may be identified with SpΣ in this way. The symmetric monoidal structure of I gives rise to a symmetric monoidal structure on IS . On morphism spaces this is given by the continuous maps ⊔ : IS (m1 , n1 ) × IS (m2 , n2 ) → IS (m1 ⊔ m2 , n1 ⊔ n2 ), that map a pair of morphisms (x, α), (y, β) to (x ∧ y, α ⊔ β). As noted in [27], this implies that the category of symmetric spectra inherits a symmetric monoidal smash product. Given symmetric spectra X and Y , this is defined by the left Kan extension, (3.1) X ∧ Y (n) = colim X(n1 ) ∧ Y (n2 ), n1 ⊔n2 →n where the colimit is over the topological category (⊔ ↓ n) of objects and morphisms in IS × IS over n. More explicitly, we may rewrite this as X ∧ Y (n) = colim α : n1 ⊔n2 →n S n−α ∧ X(n1 ) ∧ Y (n2 ), where the colimit is now over the discrete category (⊔ ↓ n) of objects and morphisms in I × I over n. Given a morphism in this category of the form α′ α → n) → (n′1 , n′2 , n′1 ⊔ n′2 −→ n), (β1 , β2 ) : (n1 , n2 , n1 ⊔ n2 − the morphism α′ specifies a homeomorphism ′ ′ ′ S n−α ∼ = S n−α ∧ S n1 −β1 ∧ S n2 −β2 , and the induced map in the diagram is defined by ′ ′ ′ S n−α ∧ X(n1 ) ∧ Y (n2 ) → S n−α ∧ S n1 −β1 ∧ X(n1 ) ∧ S n2 −β2 ∧ Y (n2 ) ′ → S n−α ∧ X(n′1 ) ∧ Y (n′2 ). The unit for the smash product is the sphere spectrum S with S(n) = S n . By definition, a symmetric ring spectrum is a monoid in this monoidal category. Spelling this out using the above notation, a symmetric ring spectrum is a symmetric spectrum X equipped with a based map 1X : S 0 → X(0) and a collection of based maps µm,n : X(m) ∧ X(n) → X(m + n), such that the usual unitality and associativity conditions hold, and such that the diagrams ′ ′ µm,m′ ◦tw ′ ′ S n−α ∧ X(m) ∧ S n −α ∧ X(m′ ) −−−−−−→ S n+n −α⊔α yα∧α′ X(n) ∧ X(n′ ) µn,n′ −−−−→ ∧ X(m + m′ ) yα⊔α′ X(n + n′ ) commute for each pair of morphisms α : m → n and α′ : m′ → n′ in I. Here ′ ′ tw flips the factors X(m) and S n −α . A ring spectrum is commutative if it Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 711 defines a commutative monoid in SpΣ . Explicitly, this means that there are commutative diagrams µm,n X(m) ∧ X(n) −−−−→ X(m + n) τm,n ytw y µn,m X(n) ∧ X(m) −−−−→ X(n + m), where the right hand vertical map is given by the left action of the (m, n)-shuffle τm,n . 3.2. Symmetric Thom spectra via I-spaces. As in Section 1.1 we write IU for the category of I-spaces. This category inherits the structure of a symmetric monoidal category from that of I in the usual way: given I-spaces X and Y , their product X ⊠ Y is defined by the Kan extension (X ⊠ Y )(n) = colim X(n1 ) × Y (n2 ), n1 ⊔n2 →n where the colimit is again over the category (⊔ ↓ n). The unit for the monoidal structure is the constant I-space I(0, −). We use term I-space monoid for a monoid in this category. This amounts to an I-space X equipped with a unit 1X ∈ X(0) and a natural transformation of I × I-diagrams µm,n : X(m) × X(n) → X(m + n) that satisfies the obvious associativity and unitality conditions. An I-space monoid X is commutative if it defines a commutative monoid in IU, that is, the diagrams µm,n X(m) × X(n) −−−−→ X(m + n) τm,n y ytw µn,m X(n) × X(m) −−−−→ X(n + m) are commutative. The family of topological monoids F (n) define a functor from I to the category of topological monoids: a morphism α : m → n in I induces a monoid homomorphism α : F (m) → F (n) by associating to an element f in F (m) the composite map n−α S ∧f Sn ∼ = Sn. = S n−α ∧ S m −−−−−→ S n−α ∧ S m ∼ ∼ S n is induced by the bijection As usual the homeomorphism S n−α ∧ S m = (n − α) ⊔ m → n specified by α and the inclusion of n − α in n. We also have the natural monoid homomorphisms (3.2) F (m) × F (n) → F (m + n), (f, g) 7→ f ∧ g defined by the usual smash product of based spaces. Applying the classifying space functor degree-wise and using that it commutes with products, we get from this the commutative I-space monoid BF : n 7→ BF (n). We write IU/BF for the category of I-spaces over BF with objects (X, f ) given by a Documenta Mathematica 14 (2009) 699–748 712 Christian Schlichtkrull map f : X → BF of I-spaces. This category inherits a symmetric monoidal structure from that of IU: given objects (X, f ) and (Y, g), the product is defined by the composition f ⊠g f ⊠ g : X ⊠ Y −→ BF ⊠ BF → BF, where the last map is the multiplication in BF . The meaning of the symbol f ⊠ g will always be clear from the context. By definition, a monoid in this monoidal structure is a pair (X, f ) given by an I-space monoid X together with a monoid morphism f : X → BF . Definition 3.3. The symmetric Thom spectrum functor T : IU/BF → SpΣ is defined by the level-wise Thom space construction T (f )(n) = T (fn ). A morphism α : m → n in I gives rise to a pullback diagram ¯ V (m) −−−−→ V (n) S n−α ∧ y y α −−−−→ BF (n) BF (m) ¯ denotes the fibre-wise smash product. On fibres this restricts to the in which ∧ homeomorphism S n−α ∧ S m → S n specified by α. Pulling this diagram back via f and applying the Thom space construction, we get the required structure maps S n−α ∧ T (fm ) ∼ = T (α ◦ fm ) → T (fn ). Notice, that this Thom spectrum functor is related to that in Section 2 by a commutative diagram of functors T IU/BF −−−−→ SpΣ y y (3.4) T N U/BF −−−−→ Sp, where the vertical arrows represent the obvious forgetful functors. Recall the notion of a strong symmetric monoidal functor from [22], Section XI.2. We now prove Theorem 1.1 stating that the symmetric Thom spectrum is strong symmetric monoidal. Proof of Theorem 1.1. It is clear that we have a canonical isomorphism S → T (∗). We must show that given objects (X, f ) and (Y, g) in IU/BF there is a natural isomorphism T (f ) ∧ T (g) ∼ = T (f ⊠ g). By definition, X ⊠ Y (n) is the colimit of the (⊔ ↓ n)-diagram α → n) 7→ X(n1 ) × Y (n2 ). (n1 , n2 , n1 ⊔ n2 − Given α : n1 ⊔ n2 → n, let α(f, g) be the composite map fn ×gn α 2 1 −−−→ BF (n1 ) × BF (n2 ) → BF (n1 + n2 ) − → BF (n). X(n1 ) × Y (n2 ) −−− Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 713 Using these structure maps we view the above (⊔ ↓ n)-diagram as a diagram over BF (n), and since the Thom space functor preserves colimits by [20], Propositions 1.1 and 1.4, we get the homeomorphism T (f ⊠ g)(n) ∼ = colim T (α(f, g)). (⊔↓n) Furthermore, since pullback commutes with topological realization and fibrewise smash products, we have an isomorphism ¯ gn∗ 2 V (n2 ) ¯ fn∗1 V (n1 )∧ α(f, g)∗ V (n) ∼ = S n−α ∧ of sectioned spaces over BF (n), hence a homeomorphism of the associated Thom spaces T (α(f, g)) ∼ = S n−α ∧ T (fn ) ∧ T (gn ). 1 2 Combining the above, we get the homeomorphism T (f ) ∧ T (g)(n) ∼ = colim S n−α ∧ T (fn1 ) ∧ T (gn2 ) ∼ = T (f ⊠ g)(n), α specifying the required isomorphism of symmetric spectra. Corollary 3.5. If X is an I-space monoid and f : X → BF a monoid morphism, then T (f ) is a symmetric ring spectrum which is commutative if X is. Recall that the tensor of an unbased space K with a symmetric spectrum X is defined by the level-wise smash product X ∧ K+ . Similarly, the tensor of K with an I-space X is defined by the level-wise product X × K. For an object (X, f ) in IU/BF , the tensor is given by (X × K, f ◦ πX ), where πX denotes the projection onto X. We refer to [6], Chapter 6, for a general discussion of tensors in enriched categories. Proposition 3.6. The symmetric Thom spectrum functor preserves colimits and tensors with unbased spaces. Proof. The first statement follows [20], Proposition 1.1 and Corollary 1.4, which combine to show that the Thom space functor preserves colimits. The second claim is that T (f ◦ πX ) is isomorphic to T (f ) ∧ K+ which follows directly from the definition. 4. Lifting space level data to I-spaces The homotopy colimit construction induces a functor (4.1) hocolim : IU/BF → U/BFhI , I (X → BF ) 7→ (XhI → BFhI ), where, given an I-space X, we write XhI for its homotopy colimit. Our first task in this section is to verify that this is the left adjoint in a Quillen equivalence. Documenta Mathematica 14 (2009) 699–748 714 Christian Schlichtkrull 4.1. The right adjoint of hocolimI . Recall first that the homotopy colimit functor IU → U has a right adjoint that to a space Y associates the I-space n 7→ Map(B(n ↓ I), Y ). Here (n ↓ I) denotes the category of objects in I under n. We refer to [7], Section XII.2.2, and [17] for the details of this adjunction. The right adjoint in turn induces a functor U : U/BFhI → IU/BF, (X, f ) 7→ (Uf (X), U (f )) by associating to a map f : X → BFhI the map of I-spaces defined by the upper row in the pullback diagram Uf (X) y U(f ) −−−−→ BF y Map(B(− ↓ I), X) −−−−→ Map(B(− ↓ I), BFhI ) The map on the right is the unit of the adjunction. It is immediate that U is right adjoint to the homotopy colimit functor in (4.1) and we shall prove in Proposition 4.5 below that the adjunction / U/BF : U (4.2) hocolim : IU/BF o I hI is a Quillen equivalence when we give U/BFhI the model structure induced by the Quillen model structure on U and IU/BF the model structure induced by the I-model structure on IU established by Sagave-Schlichtkrull [33]. Before describing the I-model structure we recall that IU has a level model structure in which the weak equivalences and fibrations are defined level-wise. Given d ≥ 0, let Fd : U → IU be the functor that to a space K associates the I-space Fd (K) = I(d, −) × K. Thus, Fd is left adjoint to the evaluation functor that takes an I-space X to X(d). The level structure on IU is cofibrantly generated with set of generating cofibrations F I = {Fd (S n−1 ) → Fd (Dn ) : d ≥ 0, n ≥ 0} obtained by applying the functors Fd to the set I of generating cofibrations for the Quillen model structure on U. By a relative cell complex in IU we understand a map X → Y that can be written as the transfinite composition of a sequence of maps X = Y0 → Y1 → Y2 → · · · → colim Yn = Y n≥0 where each map Yn → Yn+1 is the pushout of a coproduct of generating cofibrations. It follows from the general theory for cofibrantly generated model categories that a cofibration in IU is a retract of a cell complex. We refer the reader to [15], Section 11.6, for a general discussion of level model structures on diagram categories. As explained in Section 1.2, the weak equivalences in the I-model structure on IU, that is, the I-equivalences, are the maps that induce weak homotopy equivalences of homotopy colimits. The cofibrations in the I-model structure are the same as for the level structure and the fibrations can be characterized Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 715 as the maps having the right lifting property with respect to acyclic cofibrations. Again, the I-model structure is cofibrantly generated with F I the set of generating cofibrations. There also is an explicit description of the generating acyclic cofibrations and the fibrations but we shall not need this here. Lemma 4.3. The adjunction in (4.2) is a Quillen adjunction Proof. We claim that hocolimI preserves cofibrations and acyclic cofibrations. By definition, hocolimI preserves weak equivalences in general and the first claim therefore implies the second. For the first claim it suffices to show that hocolimI takes the generating cofibrations for IU to cofibrations in U. The homotopy colimit of a map Fd (S n−1 ) → Fd (Dn ) may be identified with the map B(d ↓ I) × S n−1 → B(d ↓ I) × Dn and the claim follows since B(d ↓ I) is a cell complex. In preparation for the proof that the above Quillen adjunction is in fact a Quillen equivalence we make some general comments on homotopy colimits of I-spaces. In general, given an I-space X, the homotopy type of XhI may be very different from that of XhN . However, if the underlying N -space is convergent, then the natural map XhN → XhI is a weak homotopy equivalence by the following lemma due to Bökstedt; see [23], Lemma 2.3.7, for a published version. Lemma 4.4 ([8]). Let X be an I-space and suppose that there exists an unbounded non-decreasing sequence of integers λm such that any morphism m → n in I induces a λm -connected map X(m) → X(n). Then the inclusion {n} → I induces a (λn − 1)-connected map X(n) → XhI for all n. The structure maps F (m) → F (n) are (m − 1)-connected by the Freudentahl suspension theorem and consequently the induced maps BF (m) → BF (n) are m-connected. Thus, the proposition applies to the I-space BF and we see that the canonical map BF (n) → BFhI is (n − 1)-connected. This map can be written as the composition BF (n) → Map(B(n ↓ I), BFhI ) → BFhI where the first map is the unit of the adjunction and the second map is defined by evaluating at the vertex represented by the initial object. Since the second map is clearly a homotopy equivalence it follows that the first map is also (n − 1)-connected. Proposition 4.5. The adjunction (4.2) is a Quillen equivalence. Proof. Given a cofibrant object f : X → BF in IU/BF and a fibrant object g : Y → BFhI in U/BFhI we must show that a morphism φ : XhI → Y of spaces over BFhI is a weak homotopy equivalence if and only if the adjoint ψ : X → Ug (Y ) is an I-equivalence of I-spaces over BF . The maps φ and ψ Documenta Mathematica 14 (2009) 699–748 716 Christian Schlichtkrull are related by the commutative diagram XhIB BB BB B φ BB ! ψhI Y / Ug (Y )hI v vv vvε v v v{ v where ε denotes the counit for the adjunction. It therefore suffices to show that ε is a weak homotopy equivalence. The assumption that (Y, g) be a fibrant object means that g is a fibration and the pullback diagram Ug (Y )(n) y −−−−→ BF (n) y Map(B(n ↓ I), Y ) −−−−→ Map(B(n ↓ I), BFhI ) used to define Ug (Y ) is therefore homotopy cartesian. By the remarks following Lemma 4.4 it follows that the vertical maps are (n − 1)-connected. The counit ε admits a factorization Ug (Y )hI → Map(B(− ↓ I), Y )hI → Y where the first map is a weak homotopy equivalence by the above discussion and the second map is a weak homotopy equivalence since B(− ↓ I) is level-wise contractible. This completes the proof. The functor U is only homotopically well-behaved when applied to fibrant objects. We define a (Hurewicz) fibrant replacement functor Γ on U/BFhI as in (2.4) (replacing BF (n) by BFhI ) and we write U ′ for the composite functor U ◦ Γ. This is up to natural homeomorphism the same as the functor obtained by evaluating the homotopy pullback instead of the pullback in the diagram defining Uf (X). 4.2. The I-space lifting functor R. As discussed in Section 1.2, the functor U does not have all the properties one may wish when constructing Thom spectra from maps to BFhI . In this section we introduce the I-space lifting functor R and we establish some of its properties. Given a space X and a map f : X → BFhI , we shall view this as a map of constant I-spaces. In order to lift it to a map with target BF , consider the I-space BF defined by BF (n) = hocolim BF ◦ πn , (I↓n) where πn : (I ↓ n) → I is the forgetful functor that maps an object m → n to m. By definition, BF is the homotopy left Kan extension of BF along the identity functor on I, see Appendix A.1. Since the identity on n is a terminal object in (I ↓ n) there results a canonical homotopy equivalence tn : BF (n) → BF (n) for each n. Lemma 4.6. The map πn : BF (n) → BFhI induced by the functor πn is (n−1)connected. Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 717 Proof. The homotopy equivalence tn has a section induced by the inclusion of the terminal object in (I ↓ n), such that the canonical map BF (n) → BFhI factors through BF (n). The result therefore follows from Lemma 4.4 and the above discussion. Consider now the diagram of I-spaces π t − BF − BFhI ← → BF, where the right hand map is the level-wise equivalence specified above and the left hand map is induced by the functors πn . Here we again view BFhI as a constant I-space. We define the I-space Rf (X) to be the level-wise homotopy pullback of the diagram of I-spaces f π − BF , X− → BFhI ← (4.7) that is, Rf (X)(n) is the space I {(x, ω, b) ∈ X × BFhI × BF (n) : ω(0) = f (x), ω(1) = π(b)}. Notice, that the two projections Rf (X) → X and Rf (X) → BF are level-wise Hurewicz fibrations of I-spaces. The functor R is defined by (4.8) t → BF ). R : U/BFhI → IU/BF, (f : X → BFhI ) 7→ (R(f ) : Rf (X) → BF − When there is no risk of confusion we write R(X) instead of Rf (X). Proposition 4.9. The I-space Rf (X) is convergent and R(f ) is level-wise T -good. Proof. Since Rf (X) is defined as a homotopy pullback, we see from Lemma 4.6 that the map Rf (X)(n) → X is (n − 1)-connected for each n, hence Rf (X) is convergent. We claim that R(f ) classifies a well-based quasifibration at each level. In order to see this we first observe that t∗ V (n) is a well-based quasifibration over BF (n) by Lemma A.4. Thus, R(f )∗ V (n) is a pullback of a well-based quasifibration along the Hurewicz fibration Rf (X)(n) → BF (n), hence is itself a well-based quasifibration by Lemma 2.2. ∼ Proposition 4.10. There is a natural level-wise equivalence Rf (X) − → Uf′ (X) over BF . Proof. Given a map f : X → BFhI , consider the diagram of I-spaces X f / BFhI o π BF t Map(B(− ↓ I), X) f / Map(B(− ↓ I), BFhI ) o BF where we view X and BFhI as constant I-spaces and the corresponding vertical maps are induced by the projection B(− ↓ I) → ∗. The left hand square is strictly commutative and we claim that the right hand square is homotopy commutative. Indeed, with notation as in Appendix A.1, BF is the homotopy Documenta Mathematica 14 (2009) 699–748 718 Christian Schlichtkrull Kan extension id h∗ BF along the identity functor on I and the adjoints of the two compositions in the diagram are the two maps hocolimI BF → BFhI shown to be homotopic in Lemma A.3. Using the canonical homotopy from that lemma we therefore get a canonical homotopy relating the two composites in the right hand square. The latter homotopy in turn gives rise to a natural maps of the associated homotopy pullbacks, that is, to a natural map Rf (X) → Uf′ (X). Since the vertical maps in the above diagram are level-wise equivalences the same holds for the map of homotopy pullbacks. Corollary 4.11. The functors R and hocolimI are homotopy inverses in the sense that there is a chain of natural weak homotopy equivalences Rf (X)hI ≃ X of spaces over BFhI and a chain of natural I-equivalences RfhI (XhI ) ≃ X of I-spaces over BF . Proof. It follows easily from Proposition 4.5 and its proof that the functor U ′ has this property and the same therefore holds for R by Proposition 4.11. The functor R has good properties both formally and homotopically. Proposition 4.12. The functor R in (4.8) takes weak homotopy equivalences over BFhI to level-wise equivalences over BF and preserves colimits and tensors with unbased spaces. Proof. The first statement follows from the homotopy invariance of homotopy pullbacks. In order to verify that R preserves colimits, we first observe that BFhI is locally equiconnected (the diagonal BFhI → BFhI × BFhI is a cofibration) by [19], Corollary 2.4. We then view Rf (X) as the pullback of X along the level-wise Hurewicz fibrant replacement Γπ (BF ) → BFhI and the result follows from [20], Propositions 1.1 and 1.2, which together state that the pullback functor along a Hurewicz fibration preserves colimits provided the base space is locally equiconnected. The last statement about preservation of tensors is the claim that if K is an unbased space and (X, f ) an object of U/BFhI , then R takes (X × K, f ◦ πX ) to Rf (X) × K; this follows immediately from the definition. Combining this result with Proposition 2.10, Proposition 3.6 and Proposition 4.9, we get the following corollary in which we define the Thom spectrum functor on U/BFhI using R. Corollary 4.13. The Thom spectrum functor (4.14) R T → IU/BF − → SpΣ T : U/BFhI − takes values in the subcategory of well-based, connective and convergent symmetric spectra. It takes weak homotopy equivalences over BFhI to level-wise equivalences and preserves colimits and tensors with unbased spaces. The functor R also behaves well with respect to cofibrations as we explain next. We follow [27] in using the term h-cofibration for a morphism having the Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 719 homotopy extension property. Thus, a map i : A → X in U is an h-cofibration if and only if the induced map from the mapping cylinder X ∪i (A × I) → X × I (4.15) admits a retraction. By our conventions, this is precisely what we mean by a cofibration of spaces in this paper. Given a base space B, a morphism i in U/B is an h-cofibration if the analogous morphism (4.15) admits a retraction in U/B; we emphasize this by saying that i is a fibre-wise h-cofibration. These conventions also apply to define h-cofibrations in IU and, given an I-space B, fibre-wise h-cofibrations in IU/B with the corresponding mapping cylinders defined level-wise. A morphism i : A → X in SpΣ is an h-cofibration if the mapping cylinder X ∪i (A ∧ I+ ) is a retract of X ∧ I+ . Proposition 4.16. The functor R takes maps over BFhI that are cofibrations in U to fibre-wise h-cofibrations in IU/BF and the Thom spectrum functor (4.14) takes such maps to h-cofibrations of symmetric spectra. Proof. Notice first that we may view Rf (X) as the pullback of BF along the fibrant replacement Γf (X) → BFhI . Given a morphism (A, f ) → (X, g) in U/BFhI such that A → X is a cofibration, the induced map Γf (A) → Γg (X) is a fibre-wise h-cofibration by [20], IX, Proposition 1.11. Since fibre-wise hcofibrations are preserved under pullback, this in turn implies that Rf (A) → Rg (X) is a fibre-wise h-cofibration over BF , hence over BF . It follows from Proposition 3.6 that the Thom spectrum functor on IU/BF takes fibre-wise hcofibrations to h-cofibrations. Combining this with the above gives the result. 4.3. Preservation of monoidal structures. Recall from [22], Section XI.2, that given monoidal categories (A, , 1A ) and (B, △, 1B ), a monoidal functor Φ : A → B is a functor Φ together with a morphism 1B → Φ(1A ) and a natural transformation Φ(X)△Φ(Y ) → Φ(XY ), satisfying the usual associativity and unitality conditions. It follows from the definition that if A is a monoid in A, then Φ(A) inherits the structure of a monoid in B. Since (unbased) homotopy colimits commute with products, we may view hocolimI as a monoidal functor IU → U with structure maps hocolim X × hocolim Y ∼ = hocolim X × Y → hocolim X ⊠ Y I I I×I I induced by the universal natural transformation X(m)×Y (n) → X ⊠Y (m⊔n) of I × I-diagrams. The unit morphism is induced by the inclusion of the initial object 0, thought of as a vertex in BI. Since BF is a monoid in IU, BFhI inherits the structure of a topological monoid. It follows that we may also view U/BFhI as a monoidal category and the following result is then clear from the definition. Proposition 4.17. The functor hocolimI in (4.1) is monoidal. Documenta Mathematica 14 (2009) 699–748 720 Christian Schlichtkrull However, the functor hocolimI is not symmetric monoidal, hence does not take commutative monoids in IU to commutative topological monoids. In particular, BFhI is not a commutative monoid which is already clear from the fact that it is not equivalent to a product of Eilenberg-Mac Lane spaces. We prove in Section 6.1 that BFhI has a canonical E∞ structure and that more generally hocolimI takes E∞ objects in IU to E∞ spaces. Proposition 4.18. The functor R in (4.8) is monoidal. Proof. By definition, R(∗)(0) is the loop space of BFhI and we let ∗ → R(∗) be the map of I-spaces that is the inclusion of the constant loop in degree 0. We must define an associative and unital natural transformation of I-spaces R(X) ⊠ R(Y ) → R(X × Y ) over BF . By the universal property of the ⊠product, this amounts to an associative and unital natural transformation of I 2 -diagrams R(X)(m) × R(Y )(n) → R(X × Y )(m ⊔ n). The domain is the homotopy pullback of the diagram X × Y → BFhI × BFhI ← BF (m) × BF (n), and the target is the homotopy pullback of the diagram X × Y → BFhI ← BF (m + n). The I-space BF inherits a monoid structure from that of BF such that π : BF → BFhI is a map of monoids. Using these structure maps, we define a map from the first diagram to the second, giving the required multiplication. Since the monoids in the monoidal category U/BFhI are precisely the topological monoids over BFhI , this has the following corollary. Corollary 4.19. If X is a topological monoid and f : X → BFhI a monoid morphism, then T (f ) is a symmetric ring spectrum. This may be reformulated as saying that the Thom spectrum functor preserves the action of the associativity operad whose kth space is the symmetric group Σk , see [28], Section 3. More generally, we show in Section 6 that T preserves all operad actions of operads that are augmented over the Barratt-Eccles operad. 4.4. Comparison with the Lewis-May Thom spectrum functor. Let as before BFN denote the colimit of BF over N . In this section we recall the Thom spectrum functor on U/BFN considered in [20], Section IX, and we relate this to our symmetric Thom spectrum functor on U/BFhI . We shall use the same notation for the I-space BF and its restriction to an N -space. The colimit functor N U/BF → U/BFN has a right adjoint, again denoted U , that to an object f : X → BFN associates the map of N -spaces defined by the Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 721 upper row in the pullback diagram U(f ) Uf (X) −−−−→ BF y y f −−−−→ BFN X where the vertical map on the right is the unit of the adjunction relating the colimit and the constant functors. Here we view X and BFN as constant N -spaces. We again write U ′ for the functor obtained by composing with the Hurewicz fibrant replacement functor Γ on U/BFN . The Thom spectrum functor considered in [20] is the composition U T → N U/BF − → Sp, U/BFN − where T is the functor from Section 2. (In the language of [20] this is the Thom prespectrum associated to f . The authors go on to define a spectrum M (f ) with the property that the adjoint structure maps are homeomorphisms, but this will not be relevant for the discussion here). The first step in the comparison to our symmetric Thom spectrum functor on U/BFhI is to relate the spaces BFN and BFhI . Consider the diagram of weak homotopy equivalences t i → BFN , − BFhN − BFhI ← where i is induced from the inclusion i : N → I and t is the canonical projection from the homotopy colimit to the colimit. The former is a weak homotopy equivalence by Lemma 4.4 and the latter is a weak homotopy equivalence since the structure maps are cofibrations. Let us choose a homotopy inverse j : BFhI → BFhN of i and a homotopy relating i ◦ j to the identity on BFhI . Here we of course use that these spaces have the homotopy type of a CW-complex. The precise formulation of the comparison will depend on these choices. Let ζ be the composite homotopy equivalence j t → BFN . → BFhN − ζ : BFhI − In general, given a map φ : B1 → B2 in U, we write φ∗ : U/B1 → U/B2 for the functor defined by post-composing with φ. Lemma 4.20. Suppose that φ and ψ are maps from B1 to B2 that are homotopic by a homotopy h : B1 × I → B2 . Then the functors φ∗ and ψ∗ from U/B1 to U/B2 are related by a chain of natural weak homotopy equivalences depending on h. Proof. Let h∗ : U/B1 → U/B2 be the functor that takes f : X → B1 to f ×I h X × I −−−→ B1 × I − → B2 . The two endpoint inclusions of X in X × I then give rise to the natural weak homotopy equivalences φ∗ → h∗ ← ψ∗ . Documenta Mathematica 14 (2009) 699–748 722 Christian Schlichtkrull Applied to the homotopy relating i ◦ j to the identity on BFhI this result gives a chain of natural weak homotopy equivalences relating the composite functor j∗ i ∗ U/BFhI U/BFhI −→ U/BFhN −→ to the identity on U/BFhI . Lemma 4.21. The two compositions in the diagram U′ U/BFhI −−−−→ IU/BF ζ ∗ y∗ yi U′ U/BFN −−−−→ N U/BF are related by a chain of natural level-wise equivalences. Proof. We shall interpolate between these functors by relating both to the N space analogue of the functor U ′ on U/BFhI . Thus, given f : X → BFhN , the diagram of N -spaces f Map(B(− ↓ N ), X) − → Map(B(− ↓ N ), BFhN ) ← − BF is related by evident chains of term-wise level equivalences to the diagrams i◦f Map(i∗ B(− ↓ I), X) −−→ Map(i∗ B(− ↓ I), BFhI ) ← − BF and t◦f X −−→ BFN ← − BF. Evaluating the homotopy pullbacks of these diagrams we get a chain of natural level-wise equivalences relating the two compositions in the diagram U/BFN o t∗ U/BFhN U′ N U/BF o i∗ / U/BFhI U′ ∗ i IU/BF. By the remarks following Lemma 4.20 we therefore get a chain of natural transformations (4.22) i∗ ◦ U ′ ∼ i∗ ◦ U ′ ◦ i∗ ◦ j∗ ∼ U ′ ◦ t∗ ◦ j∗ ∼ U ′ ◦ ζ∗ , each of which is a level-wise weak homotopy equivalence. We can now compare our symmetric Thom spectrum functor to the LewisMay Thom spectrum functor on U/BFN . Since the functors T R and T U ′ on U/BFhI are level-wise equivalent by Proposition 4.10, it suffices to consider T U ′. Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 723 Proposition 4.23. The two compositions in the diagram T U′ U/BFhI −−−−→ SpΣ ζ y y∗ T U′ U/BFN −−−−→ Sp are related by a chain of level-wise equivalences. Proof. The diagram in question is obtained by composing the diagram in Proposition 4.21 with the commutative diagram (3.4). Since the chain of weak homotopy equivalences in (4.22) is contained in the full subcategory of levelwise T -good objects in N U/BF , applying T gives a chain of level-wise equivalences. 5. Homotopy invariance of symmetric Thom spectra In this section we prove the homotopy invariance result stated in Theorem 1.4 and we show how the proof can be modified to give the N -space analogue in Theorem 2.11. As for the Thom space functor, the symmetric Thom spectrum functor is not homotopically well-behaved on the whole domain category IU/BF . We define a level-wise Hurewicz fibrant replacement functor on IU/BF by applying the functor Γ in (2.4) at each level. Definition 5.1. An object (X, f ) in IU/BF is T -good if the canonical map T (f ) → T (Γ(f )) is a stable equivalence (a weak equivalence in the stable model structure) of symmetric spectra. As before we say that (X, f ) is level-wise T -good if T (f ) → T (Γ(f )) is a levelwise equivalence. The first step in the proof of Theorem 1.4 is to generalize the definition of BF to any I-space X by associating to X the I-space X defined by X(n) = hocolim X ◦ πn . (I↓n) We then have a diagram of I-spaces π t → X, −X− XhI ← where we view XhI as a constant I-space. The map t is a level-wise equivalence and π is an I-equivalence by Lemma A.2. If f : X → BF is a map of I-spaces, then we have a commutative diagram f¯ X −−−−→ π y fhI BF π y XhI −−−−→ BFhI , hence there is an induced morphism (5.2) (X, t ◦ f¯) → (RfhI (XhI ), R(fhI )) of I-spaces over BF . Documenta Mathematica 14 (2009) 699–748 724 Christian Schlichtkrull Proposition 5.3. Applying T ◦ Γ to the morphism (5.2) gives a stable equivalence of symmetric spectra. In order to prove this proposition we shall make use of the level model structure on IU recalled in Section 4.1. Let Fd : U → IU be the functor defined in that section and let us write Fd (u) : Fd (K) → BF for the map of I-spaces associated to a map of spaces u : K → BF (d). Lemma 5.4. If u is a Hurewicz fibration, then Fd (u) is level-wise T -good. Proof. The pullback of V (n) along Fd (u) is isomorphic to the coproduct of the ¯ V (d) over BF (d), where pullbacks along u of the fibre-wise suspensions S n−α ∧ α runs through the injective maps d → n. These are well-based quasifibrations by Proposition 2.1 and since u is a fibration, the same holds for the pullbacks by Lemma 2.2 and the claim follows. The idea is to first prove Proposition 5.3 for objects of the form Fd (u). Lemma 5.5. Applying T ◦ Γ to the map of I-spaces Fd (K) → R(Fd (K)hI ) over BF gives a stable equivalence of symmetric spectra. The proof of this requires some preparation. We view K as a space over BFhI via the map (5.6) ũ : K → BF (d) → BFhI , where the second map is induced by the inclusion of {d} in I. Lemma 5.7. There is a weak homotopy equivalence K → Fd (K)hI of spaces over BFhI . Proof. By definition of the homotopy colimit we may identify Fd (K)hI with B(d ↓ I) × K, where (d ↓ I) is the category of objects in I under d. Since this category has an initial object its classifying space is contractible and the result follows. In the case of the I-space Fd (K), the level-wise equivalence t : Fd (K) → Fd (K) has a section induced by the canonical map K → Fd (K)(d). Using this, we get a commutative diagram in IU/BF , ∼ (5.8) Fd (K) −−−−→ y ∼ Fd (K) y R(K) −−−−→ R(Fd (K)hI ). The upper horizontal map is a level-wise equivalence since t is and the lower horizontal map is a level-wise equivalence by the above lemma. Thus, in order to prove Lemma 5.5, we may equally well consider the vertical map on the left hand side of the diagram. Given a based space T , let FdS (T ) be the symmetric spectrum IS (d, −) ∧ T . The functor FdS so defined is left adjoint to the functor SpΣ → T that takes a symmetric spectrum to its dth space, see [27]. In particular it follows from the Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 725 definition that F0S (T ) is the suspension spectrum of T . Notice also that the Thom spectrum T (Fd (u)) associated to Fd (u) may be identified with FdS (T (u)), where as usual T (u) denotes the Thom space of the map u. Let T (ũ) be the symmetric Thom spectrum of the map ũ in (5.6) and let ΣdL T (ũ) be the left shift by d, that is, the composition of T (ũ) with the concatenation functor IS → IS , n 7→ d ⊔ n. Thus, the nth space of ΣdL T (ũ) is T (ũ)(d ⊔ n) with Σn acting via the inclusion Σn → Σd+n induced by n 7→ d ⊔ n. The condition that u be a Hurewicz fibration in the following lemma is unnecessarily restrictive, but the present formulation is sufficient for our purposes. Lemma 5.9. If u is a Hurewicz fibration, then the canonical map of spaces T (u) → T (ũ)(d) induces a π∗ -isomorphism F0S (T (u)) → ΣdL T (ũ). Proof. In spectrum degree n this is the map of Thom spaces induced by the map K → Rũ (K)(d) → Rũ (K)(d ⊔ n), viewed as a map of T -good spaces over BF (d + n). This is also a map of spaces over K via the projection Rũ (K) → K, and it therefore follows from the proof of Proposition 4.9 that its connectivity tends to infinity with n. The result then follows from Lemma 2.6. We shall prove Lemma 5.5 using the detection functor D from [37]. We recall that this functor associates to a symmetric spectrum T the symmetric spectrum DT whose nth space is the based homotopy colimit DT (n) = hocolim Ωm (T (m) ∧ S n ). m∈I By [37], Theorem 3.1.2, a map of (level-wise well-based) symmetric spectra T → T ′ is a stable equivalence if and only if the induced map DT → DT ′ is a π∗ -isomorphism. There is a closely related functor T 7→ M T , where M T is the symmetric spectrum with nth space M T (n) = hocolim Ωm (T (m ⊔ n)). m∈I Thus, M T is the homotopy colimit of the I-diagram of symmetric spectra m 7→ Ωm (Σm L T ). There is a canonical map DT → M T , which is a level-wise equivalence if T is convergent and level-wise well-based. Proof of Lemma 5.5. We claim that applying T ◦ Γ to the vertical map on the left hand side of (5.8) gives a stable equivalence, and for this we may assume without loss of generality that u is a Hurewicz fibration. Then Fd (u) is T -good by Lemma 5.4 and since R(ũ) is T -good by Proposition 4.9, it suffices to show that T (Fd (u)) → T (ũ) is a stable equivalence. Furthermore, by [27], Theorem 8.12, it is enough to show that this map is a stable equivalence after smashing with S d and by the above remarks this in turn follows if applying D gives a π∗ -isomorphism. We identify T (Fd (u)) with FdS (T (u)) and claim that there is Documenta Mathematica 14 (2009) 699–748 726 Christian Schlichtkrull a commutative diagram F0S (T (u)) ∼ y ∼ −−−−→ ΣdL T (ũ) ∼ −−−−→ Ωd (S d ∧ ΣdL T (ũ)) ∼ y ∼ D(S d ∧ FdS (T (u))) −−−−→ D(S d ∧ T (ũ)) −−−−→ M (S d ∧ T (ũ)), where the maps are π∗ -isomorphisms as indicated. The vertical map on the left hand side is induced by the space-level map T (u) → FdS (T (u))(d) → Ωd (S d ∧ FdS (T (u))(d)) → D(S d ∧ FdS (T (u)))(0). It is a fundamental property of the model structure on SpΣ that the induced map of symmetric spectra is a π∗ -isomorphism, see the proof of [37], Lemma 3.2.5. The first map in the upper row is the stable equivalence from Lemma 5.9, and the remaining indicated arrows are π∗ -isomorphisms since T (ũ) is connective and convergent. This proves the claim. We now wish to prove Proposition 5.3 by an inductive argument based on the filtration (5.10) ∅ = X0 → X1 → X2 → · · · → colim Xn = X n of a cell complex X in IU, cf. the discussion of the level model structure in Section 4.1. In order to carry out the induction step, we need to ensure that the induced maps of Thom spectra are h-cofibrations in the sense of Section 4.2. The following is the I-space analogue of [20], IX, Lemma 1.9 and Proposition 1.11. The proof is essentially the same as in the space-level case. Proposition 5.11. The functor Γ on IU/BF preserves colimits and takes morphisms in IU/BF that are h-cofibrations in IU to fibre-wise h-cofibrations. Since the symmetric Thom spectrum functor on IU/BF preserves colimits and takes fibre-wise h-cofibrations to h-cofibrations by Proposition 3.6, this has the following consequence. Proposition 5.12. The composite functor T ◦ Γ preserves colimits and takes morphisms in IU/BF that are h-cofibrations in IU to h-cofibrations of symmetric spectra. Proof of Proposition 5.3. Using the level model structure we may choose a cofibrant I-space X ′ and a level-wise equivalence X ′ → X, hence it suffices to consider the case where X is a cofibrant I-space. Then X is a retract of a cell complex which we may view as a cell complex over BF via the retraction. By functoriality we are thus reduced to the case where X is a cell complex with a filtration by h-cofibrations as in (5.10). In order to handle this case we use that both functors in (5.2) preserve colimits and tensors with unbased spaces, hence they also preserve (not necessarily fibre-wise) h-cofibrations. Applying the functor T ◦ Γ, we see that both functors in the proposition preserve colimits and take h-cofibrations of I-spaces over BF to h-cofibrations of symmetric Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 727 spectra. We prove by induction that the result holds for each of the I-spaces Xn in the filtration. By definition, Xn+1 is a pushout of a diagram of the form B ← A → Xn , where A → B is a coproduct of generating cofibrations, hence in particular an h-cofibration. We view this as a diagram of I-spaces over BF via the inclusion of Xn+1 in X and get a diagram of Thom spectra T Γ(X n ) y ←−−−− T Γ(A) y −−−−→ T Γ(B) y T Γ(R((Xn )hI )) ←−−−− T Γ(R(AhI )) −−−−→ T Γ(R(BhI )), such that the map for Xn+1 is the induced map of pushouts. By the above discussion it follows that the horizontal maps on the right hand side of the diagram are h-cofibrations and the vertical maps are stable equivalences by Lemma 5.5 and the induction hypothesis. Consequently the map of pushouts is also a stable equivalence, see [27], Theorem 8.12. Proof of Theorem 1.4. We prove that applying the functor T ◦ Γ to an Iequivalence X → Y over BF gives a stable equivalence of symmetric spectra. Consider the commutative diagram X ←−−−− y X −−−−→ R(XhI ) y y Y ←−−−− Y −−−−→ R(YhI ) of I-spaces over BF . Applying T ◦ Γ to this diagram we get a diagram of symmetric spectra where the horizontal maps are stable equivalence by Proposition 5.3 and the fact that T ◦ Γ preserves level-wise equivalences. The result now follows from Corollary 4.13 which ensures that the map R(XhI ) → R(YhI ) induces a stable equivalence. Notice, that as a consequence of the theorem, the composite functor T ◦ Γ is a homotopy functor on IU/BF in the sense that it takes I-equivalences to stable equivalences. 5.1. The proof of Theorem of 2.11. The proof of Theorem 2.11 is similar to but simpler than the proof of Theorem 1.4. We first introduce a functor RN : U/BFhN → N U/BF, which is the N -space analogue of the functor R. Let us temporarily write BF for the homotopy Kan extension of the N -space BF along the identity functor of N , that is, BF (n) = hocolim BF ◦ πn , (N ↓n) where πn is the forgetful functor (N ↓ n) → N . Given a map f : X → BFhN , we define RfN (X) to be the level-wise homotopy pullback of the diagram of N -spaces f π − BF , X− → BFhN ← Documenta Mathematica 14 (2009) 699–748 728 Christian Schlichtkrull and we define RN (f ) to be the composite map of N -spaces t RN (f ) : RfN (X) → BF − → BF. Exactly as in the I-space case there is a map of N -spaces (5.13) (X, t ◦ f ) → (RfNhN (XhN ), RN (fhN )) over BF , where we again use the (temporary) notation X for the homotopy Kan extension along the identity on N . Theorem 2.11 then follows from the following proposition in the same way that Theorem 1.4 follows from Proposition 5.3. Proposition 5.14. Applying T ◦ Γ to (5.13) gives a stable equivalence of spectra. In order to prove this we first consider the N -spaces Fd (K) defined by N (d, −) × K, where d is an object in N and K is a space. Given a map u : K → BF (d), we have the following N -space analogue of (5.8), Fd (K) −−−−→ y Fd (K) y R(K) −−−−→ R(Fd (K)hN ). However, in contrast to the I-space setting, this is a diagram of convergent N -spaces and the connectivity of the maps in degree n tends to infinity with n. Thus, the N -space analogue of Lemma 5.5 holds with a simpler proof. Proof of Proposition 5.14. We use that N U has a cofibrantly generated level model structure and as in the I-space case we reduce to the case of a cell complex. Using that the functors in (5.13) preserve colimits and h-cofibrations, the inductive argument used in the proof of Proposition 5.3 then also applies in the N -space setting. 6. Preservation of operad actions Let C be an operad as defined in [28] and notice that C defines a monad C on the symmetric monoidal category IU in the usual way by letting ∞ a C(k) ×Σk X ⊠k . C(X) = k=0 We define a C-I-space to be an algebra for this monad and write IU[C] for the category of such algebras. More explicitly, a C-I-space is an I-space X together with a sequence of maps of I-spaces θk : C(k) × X ⊠k → X, satisfying the associativity, unitality and equivariance relations listed in [28], Lemma 1.4. By the universal property of the ⊠-product, θk is determined by a natural transformation of I k -diagrams (6.1) θk : C(k) × X(n1 ) × · · · × X(nk ) → X(n1 + · · · + nk ) Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 729 and the equivariance condition amounts to the commutativity of the diagram θk ◦(σ×id) C(k) × X(n1 ) × · · · × X(nk ) yid×σ −−−−−−→ C(k) × X(nσ−1 (1) ) × · · · × X(nσ−1 (k) ) θ −−−k−→ X(n1 + · · · + nk ) σ(n ,...,n ) k ∗ y 1 X(nσ−1 (1) + · · · + nσ−1 (k) ) for all elements σ in Σk . Here σ permutes the factors on the left hand side of the diagram and σ(n1 , . . . , nk ) denotes the permutation of n1 ⊔ · · · ⊔ nk that permutes the k summands as σ permutes the elements of k. As defined in [28], the 0th space of C is a one-point space, so that θ0 specifies a base point of X. Notice, that an action of the one-point operad ∗ on an I-space X is the same thing as a commutative monoid structure on X. In this case the projection C → ∗ induces a C-action on X for any operad C. This applies in particular to the commutative I-space monoid BF . In similar fashion an operad C defines a monad C on the category SpΣ by letting ∞ _ C(X) = C(k)+ ∧Σk X ∧k k=0 Σ and we write Sp [C] for the category of algebras for this monad. Thus, an object of SpΣ [C] is a symmetric spectrum X together with a sequence of maps of symmetric spectra θk : C(k)+ ∧ X ∧k → X, satisfying the analogous associativity, unitality and equivariance relations. By the universal property of the smash product, θk is determined by a natural transformation of ISk -diagrams, (6.2) θk : C(k)+ ∧ X(n1 ) ∧ · · · ∧ X(nk ) → X(n1 + · · · + nk ). The naturality condition can be formulated explicitly as follows. Given a family of morphisms αi : mi → ni in I for i = 1, . . . , k, let α = α1 ⊔ · · · ⊔ αk . Writing n = n1 + · · · + nk and making the identification S n1 −α1 ∧ · · · ∧ S nk −αk = S n−α , we require that the diagram S n−α ∧θ k S n−α ∧ C(k)+ ∧ X(m1 ) ∧ · · · ∧ X(mk ) −−−−−−→ S n−α ∧ X(m1 + · · · + mk ) y y C(k)+ ∧ X(n1 ) ∧ · · · ∧ X(nk ) θ −−−k−→ X(n1 + · · · + nk ) be commutative. We now show that the symmetric Thom spectrum functor behaves well with respect to operad actions. Given an operad C and a map of I-spaces f : X → BF , let C(f ) be the composite map C(f ) : C(X) → C(BF ) → BF. Documenta Mathematica 14 (2009) 699–748 730 Christian Schlichtkrull The following is the analogue in our setting of [20], Theorem IX 7.1. It is a formal consequence of the fact that T is a strong symmetric monoidal functor that preserves colimits and tensors with unbased spaces. Proposition 6.3. There is a canonical isomorphism of symmetric spectra T (C(f )) = C(T (f )). Corollary 6.4. The Thom spectrum functor on IU/BF preserves operad actions in the sense that there is an induced functor T : IU[C]/BF → SpΣ [C]. 6.1. Operad actions preserved by hocolimI . As in Section 1.2 we use the notation E for the Barratt-Eccles operad. We recall that the kth space E(k) is the classifying space of the translation category Σ̃k that has the elements of Σk as its objects. A morphism ρ : σ → τ in Σ̃k is an element ρ ∈ Σk such that ρσ = τ ; see [29], Section 4 (but notice that the order of the composition in Σ̃k is defined differently here). In the following proposition, C denotes an arbitrary operad and E × C denotes the product operad whose kth space is the product E(k) × C(k). Proposition 6.5. The functor hocolimI induces a functor hocolim : IU[C] → U[E × C]. I Proof. Let I(X) be the topological category whose space of objects is the disjoint union of the spaces X(n) indexed by the objects n in I, and in which a morphism (m, x) → (n, y) is specified by a morphism α : m → n in I such that α∗ (x) = y. Then it follows from the definition of the homotopy colimit that XhI may be identified with the classifying space BI(X); see Appendix A for details. In the following we shall view the spaces C(k) as topological categories with only identity morphisms. For each k, consider the functor of topological categories ψk : Σ̃k × C(k) × I(X)k → I(X), that maps a tuple of objects σ ∈ Σk , c ∈ C(k), and (n1 , x1 ), . . . , (nk , xk ), to (nσ−1 (1) ⊔ · · · ⊔ nσ−1 (k) , σ(n1 , . . . , nk )∗ θk (c, x1 , . . . , xk )). Here θk denotes the C(k)-action on X and σ(n1 , . . . , nk ) is defined as at the beginning of this section. If ρ : σ → τ is a morphism in Σ̃k , then the induced morphism in I(X) is specified by ψk (ρ) = ρ(nσ−1 (1) , . . . , nσ−1 (k) ), and if α ~ denotes a k-tuple of morphisms in I(X) whose ith component is specified by αi : ni → mi , then the induced morphism in I(X) is specified by ψk (~ α) = ασ−1 (1) ⊔ · · · ⊔ ασ−1 (k) . Since the classifying space functor preserves products, these functors give rise to maps ψk : E(k) × C(k) × BI(X)k → BI(X), Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 731 and it is straightforward to check that this defines an E × C-action on BI(X). The associativity, unitality, and equivariance conditions may all be checked on the categorical level. Letting C be the commutativity operad ∗, it follows in particular that if X is a commutative I-space monoid, then XhI inherits an E-action. In this case the action is induced by a permutative structure on the category I(X) introduced in the above proof, cf. [29], Section 4. This applies in particular to the I-space BF giving an E-action on BFhI . We say that an operad C is augmented over E if there is a specified morphism of operads C → E. In this case we may restrict an (E × C)-action to the diagonal C-action via the morphism C → E × C. Corollary 6.6. If C is augmented over E, then hocolimI induces a functor hocolim : IU[C]/BF → U[C]/BFhI . I 6.2. Operad actions preserved by R. In order to prove that the I-space lifting functor R preserves operad actions, we need the following lemma in which we view BFhI as a constant E-I-space. Lemma 6.7. The I-space BF has an E-action such that π : BF → BFhI is a morphism of E-I-spaces. Proof. Consider more generally a commutative I-space monoid X, and let X be the I-space defined in Section 5. For each object n in I, let I/n(X) be the topological category whose classifying space is X(n). Thus, the object space is given by a X(m), α : m→n where the coproduct is over the objects in (I ↓ n); see Appendix A for details. Consider for each k the functor ψk : Σ̃k × I/n1 (X) × · · · × I/nk (X) → I/(n1 ⊔ · · · ⊔ nk )(X) that maps a tuple of objects σ in Σ̃k and (αi , xi ) in I/ni (X) for i = 1, . . . , k, to the object (α, xσ−1 (1) . . . xσ−1 (k) ), where α is the morphism ασ−1 (1) ⊔···⊔ασ−1 (k) α : mσ−1 (1) ⊔ · · · ⊔ mσ−1 (k) −−−−−−−−−−−−−→ nσ−1 (1) ⊔ · · · ⊔ nσ−1 (k) σ−1 (nσ−1 (1) ,...,nσ−1 (k) ) −−−−−−−−−−−−−−−−→ n1 ⊔ · · · ⊔ nk and the second factor is the product of the elements xσ−1 (1) , . . . , xσ−1 (k) using the monoid structure. The induced maps of classifying spaces ψk : E(k) × X(n1 ) × · · · × X(nk ) → X(n1 + · · · + nk ) then specify the required E-action on X. With this definition it is clear that the canonical morphism X → XhI is a morphism of E-I-spaces. Documenta Mathematica 14 (2009) 699–748 732 Christian Schlichtkrull Proposition 6.8. Let C be an operad augmented over the Barratt-Eccles operad. Then the I-space lifting functor R induces a functor R : U[C]/BFhI → IU[C]/BF. Proof. We give BFhI the C-action defined by the augmentation to E. Let f : X → BFhI be a map of C-spaces and consider the diagram f π − BF X− → BFhI ← defining Rf (X). Pulling the E-action on BF defined in Lemma 6.7 back to a Caction, this is a diagram of C-I-spaces. Thus, Rf (X) is a homotopy pullback of a diagram in IU[C], hence is itself an object in this category and the projections Rf (X) → X and Rf (X) → BF are maps of C-I-spaces, see [28], Section 1. Since the equivalence t : BF → BF is also a map of C-I-spaces, the conclusion follows. Combining this with Corollary 6.4 we get the following. Corollary 6.9. If C is an operad that is augmented over E, then the Thom spectrum functor on U/BFhI induces a functor T : U[C]/BFhI → SpΣ [C]. 7. The Thom isomorphism Let M F be the symmetric Thom spectrum associated to the identity BF → BF , and let M SF be the symmetric Thom spectrum associated to the inclusion BSF → BF . Here SF (n) denotes the submonoid of orientation preserving based homotopy equivalences (those that are homotopic to the identity) and BSF is the corresponding commutative I-space monoid. We first construct canonical orientations of these Thom spectra, and for this we need convenient models of Eilenberg-Mac Lane spectra. 7.1. Eilenberg-Mac Lane spectra and orientations. Let A be a discrete ring, and write A[−] for the functor that to a topological space X associates the free topological A-module A[X] generated by X, see e.g. [42], Section 2.3. In the special case where X is the realization of a simplicial set X• this may be identified with the realization of the simplicial A-module A[X• ]. If X is based, we write A(X) for the topological A-module A[X]/A[∗]. The functor A(−) defined in this way is left adjoint to the forgetful functor from topological A-modules to based spaces. It is well-known that A(S n ) is a model of the Eilenberg-Mac Lane space K(A, n), and that when equipped with the obvious structure maps this defines a model of the Eilenberg-Mac Lane spectrum for A as a symmetric ring spectrum. In order to define the orientations, we shall consider a variant of this construction. Let FA (n) be the topological monoid of continuous A-linear endomorphisms of A(S n ) and notice that by the above remarks this is homotopy equivalent to A considered as a discrete multiplicative monoid. Writing SFA (n) for the connected component corresponding to Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 733 the unit of A, this is then a contractible topological monoid. Applying the bar construction as in Section 2, we get a well-based quasifibration B(∗, SFA (n), A(S n )) → BSFA (n), and we define the Eilenberg-Mac Lane spectrum HA to be the symmetric spectrum with nth space HA(n) = B(∗, SFA (n), A(S n ))/BSFA (n). It is easy to check that this is a commutative symmetric ring spectrum which is level-wise equivalent to the usual model for the Eilenberg-Mac Lane spectrum considered above. Since HA is flat in the sense of [4], the functor HA ∧ (−) preserves stable equivalences between well-based spectra; this follows from a slight refinement of the argument used in [4]. (Alternatively, one can check that the arguments in [35], Proposition 5.14, works equally well with Quillen cofibrations of spaces replaced by our notion of (h-)cofibrations.) Let now A = Z/2 and observe that the functor Z/2(−) defines a map of sectioned quasifibrations B(∗, F (n), S n ) → B(∗, SFZ/2 (n), Z/2(n)). The canonical orientation of M F is the induced map of commutative symmetric ring spectra M F → HZ/2. Similarly, the functor Z(−) defines a map of sectioned quasifibrations B(∗, SF (n), S n ) → B(∗, SFZ (n), Z(S n )) and the canonical orientation of M SF is the induced map of commutative symmetric ring spectra M SF → HZ. 7.2. The Thom isomorphism. We first consider the Thom isomorphism with Z/2-coefficients. Given a map f : X → BFhI , the I-space lift Rf (X) → BF induces a map of symmetric spectra T (f ) → M F , and we define the HZ/2orientation of T (f ) to be the composition T (f ) → M F → HZ/2. As explained in Section 1.4, the orientation induces a map of symmetric spectra (7.1) T (f ) ∧ HZ/2 → X+ ∧ HZ/2. Since our construction of the Thom spectrum functor has good properties both formally and homotopically, the proof that this is a stable equivalence is almost completely formal. Theorem 7.2. The map of symmetric spectra (7.1) is a stable equivalence. Proof. Both functor in the theorem are homotopy functors on U/BFhI in the sense that they take weak homotopy equivalences to stable equivalences; this follows from Corollary 4.13 and the fact that HZ/2 is flat. Thus, we may assume that X is a CW-complex and consider the filtration of X by skeleta Documenta Mathematica 14 (2009) 699–748 734 Christian Schlichtkrull X n such that X −1 is the empty set and X n is homeomorphic to the pushout of a diagram of the form a a X n−1 ← S n−1 → Dn . Since both functors in the theorem preserve pushouts and h-cofibrations, it suffices by [27], Theorem 8.12, to consider the case where the domain of f is of the form Dn or S n . If f is the inclusion of the basepoint ∗ → BFhI , then the unit of the I-space monoid Rf (∗) gives a stable equivalence S → T (f ) and the composition ∼ S ∧ HZ/2 − → T (f ) ∧ HZ/2 → ∗+ ∧ HZ/2 is the identity on HZ/2. Using the homotopy invariance of the Thom spectrum functor, this easily implies the result for Dn . Identifying S n with the pushout of the diagram Dn ← S n−1 → Dn , the result for S n then follows by an inductive argument. 7.3. The integral Thom isomorphism. Using the commutative I-space monoid BSF instead of BF , we get a a monoidal I-space lifting functor R : U/BSFhI → IU/BSF defined in analogy with the I-space lifting functor on U/BFhI . The two lifting functors are related by a diagram R U/BSFhI −−−−→ IU/BSF y y R U/BFhI −−−−→ IU/BF, which is commutative up to natural I-equivalence. Thus, the two natural ways to define a Thom spectrum functor on U/BSFhI are equivalent up to stable equivalence. For the definition of orientations it is most convenient to define the Thom spectrum functor on U/BSFhI to be the composition R T → IU/BSF → IU/BF − → SpΣ . T : U/BSFhI − With this definition we have a canonical integral orientation of the Thom spectrum associated to a map X → BSFhI , defined by the composition T (f ) → M SF → HZ. The orientation again gives rise to a map of symmetric spectra (7.3) T (f ) ∧ HZ → X+ ∧ HZ and the proof of the integral version of the Thom isomorphism theorem is completely analogous to the HZ/2-version. Theorem 7.4. The map (7.3) is a stable equivalence. We can now verify the claim in Theorem 1.8 that the Thom equivalence is strictly multiplicative. Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 735 Proof of Theorem 1.8. Let H denote either one of the commutative symmetric ring spectra HZ/2 or HZ, and view H as an object in SpΣ [C] by projecting C onto the commutativity operad. We claim that (1.6) is a diagram in SpΣ [C] when we give each of the terms the diagonal C-action. For the first two maps this follows from Proposition 6.8 and Corollary 6.9, which imply that the Thom diagonal and T (f ) → M F (or T (f ) → M SF ) are both C-maps. For the last map the claim follows from the fact that the multiplication H ∧ H → H is a map of commutative symmetric ring spectra, hence in particular a C-map. 8. Symmetrization of diagram Thom spectra In this section we first generalize the definition of the symmetric Thom spectrum functor to other types of diagram spectra. We then show how the results in the previous sections can be used to turn such diagram Thom spectra into symmetric spectra. 8.1. Diagram spaces and diagram Thom spectra. Given a small category D, we define a D-space to be a functor X : D → U and we write DU for the category of such functors. Suppose that we are given a functor φ : D → I. Then we can generalize the notion of a symmetric spectrum by introducing the topological category DS that has the same objects as D, but whose morphism spaces are defined by _ S b−α , DS (a, b) = α∈D(a,b) b−α where S is shorthand notation for S φ(b)−φ(α) , cf. Section 3.1. The composition is defined as for IS . We define a D-spectrum to be a continuous based functor DS → T and we write DS T for the category of such functors. Thus, a D-spectrum is given by a family of based spaces X(a) indexed by the objects a in D, together with a family of based structure maps S b−α ∧ X(a) → X(b) indexed by the morphisms α : a → b in D. It is required (i) that the structure map associated to an identity morphism 1a : a → a is the canonical identification S 0 ∧ X(a) → X(a), and (ii) that given a pair of composable morphisms α : a → b and β : b → c, the following diagram is commutative S c−β ∧ S b−α ∧ X(a) −−−−→ S c−β ∧ X(b) y y S c−βα ∧ X(a) −−−−→ X(c). In particular, if φ denotes the identity functor on I, then IS T is an alternative notation for the category of symmetric spectra. Suppose now that D has the structure of a strict monoidal category. As in the case of I-spaces, DU inherits a monoidal structure from D which is symmetric monoidal if D is. If in addition φ is strict monoidal, then the monoidal structure of D also induces a monoidal structure on DS which is symmetric monoidal if D and φ are. This in turn induces a monoidal structure on the category of D-spectra DS T which again is symmetric monoidal if D and φ are. The I-space BF pulls back to a D-space Documenta Mathematica 14 (2009) 699–748 736 Christian Schlichtkrull via φ and the definition of the symmetric Thom spectrum functor immediately generalizes to give a Thom spectrum functor T : DU/BF → DS T . The proof of Theorem 1.1 generalizes to show that this is a strong monoidal functor which is symmetric monoidal if D and φ are. 8.2. Examples of diagram Thom spectra. Many examples of Thom spectra arise from compatible families of groups over the topological monoids F (n). It often happens that such a family defines a D-diagram of groups for some strict monoidal category D over I and if the induced maps of classifying spaces define a D-space over BF we get an associated D-Thom spectrum. We begin by fixing notation for some of the relevant categories. For each k ≥ 1 we have the strict symmetric monoidal faithful functor ψk : I → I, n 7→ k · · ⊔ k} . | ⊔ ·{z n and we write I[k] for its image in I. Thus, I[k] is a strict symmetric monoidal category whose objects have cardinality a multiple of k, and whose morphisms permute blocks of k letters simultaneously. Let us write M for the subcategory of injective order preserving morphisms in I. This inherits a strict monoidal (but not symmetric monoidal) structure from I and we similarly define monoidal subcategories M[k] for k ≥ 1. Example 8.1 (The classical groups). The orthogonal groups O(n) and the special orthogonal groups SO(n) define the commutative I-space monoids BO and BSO that give rise to the commutative symmetric ring spectra M O and M SO. The unitary groups U (n) and the special unitary groups SU (n) define the commutative I[2]-space monoids BU and BSU that give rise to the commutative I[2]-ring spectra M U and M SU . The symplectic groups Sp(n) define the commutative I[4]-space monoid BSp that gives rise to the commutative I[4]-spectrum M Sp. Example 8.2 (Discrete groups and I-spaces). The symmetric groups Σn define the commutative I-space monoid BΣ in which the monoid structure is induced by concatenation of permutations. This gives rise to the commutative symmetric ring spectrum M Σ whose associated bordism theory has been studied by Bullett [9]. Other systems of discrete groups that give rise to symmetric ring spectra include the general linear groups GLn (Z), the groups (Z/2)n of diagonal matrices with entries ±1, and the groups Σn ≀ Z/2 of permutation matrices with entries ±1. For details and more examples, see [9] and [12]. Example 8.3 (Braid groups and M-spaces). The family of braid groups B(n) defines an M-space monoid BB in a natural way. We refer to [5] for the definition and basic properties of the braid groups. If we view an element of B(n) as a system of n strings in the usual way, then the monoid structure on Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 737 BB is induced by concatenation of such systems. Let ρ denote the sequence of monoid homomorphisms ρn : B(n) → Σn → F (n) where the first map takes a system of strings to the induced permutation of the endpoints and the second map in the canonical inclusion. This defines a map Bρ : BB → BF of M-space monoids and we write M B for the associated MThom ring spectrum. The underlying spectrum of M B has been analyzed in [9] and [10], where it is shown to be equivalent to the Eilenberg-Mac Lane spectrum HZ/2. Suppose that G is an M-diagram of groups over the monoids F (n) and that the homomorphisms ρn can be factored as B(n) → G(n) → F (n). If the M-space BG admits a monoid structure such that the induced map BB → BG is a map of M-space monoids over BF it then follows that M G is an M-module spectrum over M B. For example, this applies to M Σ and M O but not to M SO. We show how to symmetrize the constructions so as to get symmetric Thom spectra in Section 8.3. Again we refer to [9], [10], [12] for further examples. Example 8.4 (Maps to BF (k) and I[k]-spaces). For our next class of examples we need some preliminary definitions. Let X be a based space and let X • be the I-space defined by n 7→ X n . Given a morphism α : m → n, the induced map α∗ : X m → X n is defined by α∗ (x1 , . . . , xm ) = (xα−1 (1) , . . . , xα−1 (n) ), with the convention that x∅ is the base point in X. We give X • the structure of a commutative I-space monoid using the identifications X m × X n = X m+n . Suppose now that f : X → BF (k) is a based map. Then we view X • as an I[k]-space via the isomorphism ψk : I → I[k], and the maps µ → BF (k · · ⊔ k}) X n → BF (k)n − | ⊔ ·{z n define a map of I[k]-space monoids X • → BF , where we view BF as an I[k]-space by restriction. We write M X ∧∞ for the associated commutative I[k]-ring spectrum, the function f being understood. In the cases X = BO(1) and X = BU (1), we get the Thom spectra M O(1)∧∞ and M U (1)∧∞ that represent the bordism theories of manifolds with stable normal bundle given as an ordered sum of real or complex line bundles. These Thom spectra have been analyzed by Arthan and Bullett [1], [9]. Letting X = BO(k) or X = BU (k), we similarly get the I[k]-spectrum M O(k)∧∞ and the I[2k]-spectrum M U (k)∧∞ . 8.3. Symmetrization of diagram Thom spectra. As demonstrated in the last section, many Thom spectra naturally arise as D-Thom spectra associated to maps of D-spaces f : X → BF for suitable monoidal categories D over I. In the applications it is often convenient to replace such a D-Thom spectrum by a symmetric Thom spectrum and our preferred way of doing is to first transform f to a map of I-spaces and then evaluate the symmetric Thom spectrum functor on this transformed map. We shall discuss two ways of performing Documenta Mathematica 14 (2009) 699–748 738 Christian Schlichtkrull this “symmetrization” procedure: in this section we consider symmetrizations using the I-space lifting functor R and in the next section we consider symmetrizations via (homotopy) Kan extension. For simplicity we shall from now on assume that D is a monoidal subcategory of I such that the intersection D ∩ N is a cofinal subcategory of N . Thus, any D-spectrum has an underlying D ∩ N -spectrum and we define the spectrum homotopy groups in the usual way by evaluating the colimit of the associated D ∩ N -diagram of homotopy groups. Given a map f : X → BF , we write (by abuse of notation) fhD for the composite map f hD fhD : XhD −− → BFhD → BFhI . Applying the I-space lifting functor R to this map we get a functor (8.5) DU/BF → IU/BF, f R(fhD ) (X − → BF ) 7→ (RfhD (XhD ) −−−−−→ BF ). We say that a map of D-spaces X → Y is a D-equivalence if the induced map XhD → YhD is a weak homotopy equivalence. Lemma 8.6. The restriction of RfhD (XhD ) to a D-space is related to X by a chain of D-equivalences over BF . Proof. In analogy with the case of I-spaces considered in Section 5, there is a diagram of D-equivalences X ← X → RfhD (XhD ) over BF . It follows from the D-space analogue of Bökstedt’s approximation Lemma 4.4 that if X → Y is a D-equivalence of convergent D-spaces X and Y , then the connectivity of the maps X(d) → Y (d) tends to infinity with d. The previous lemma therefore has the following consequence. Proposition 8.7. If X is a convergent D-space and f : X → BF is a map of D-spaces which is level-wise T -good, then the restriction of the symmetric spectrum T (R(fhD )) to a D-spectrum is π∗ -equivalent to T (f ). This construction preserves multiplicative structures in the sense that if X is a D-space monoid and f : X → BF a map of D-space monoids, then fhD is a map of topological monoids and T (fhD ) is a symmetric ring spectra by Lemma 4.19. In the following we consider the effect of applying the construction to the examples considered in the previous section. Example 8.8. Let X be an I[k]-space with an action of an operad C that is augmented over the Barratt-Eccles operad. If f : X → BF is a map of C-I[k]spaces, then the induced map fhI[k] : XhI[k] → BFhI[k] → BFhI is map of C-spaces and it follows from Corollary 6.9 that the symmetric spectrum T (fhI[k] ) inherits a C-action. Here we use the canonical isomorphism of categories ψk : I → I[k] to identify XhI[k] with (ψk∗ X)hI , and we transfer the C-action on (ψk∗ X)hI defined in Corollary 6.6 to XhI[k] via this identification. This applies in particular to the map of commutative I[2]-spaces BU → BF Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 739 to give a model of the Thom spectrum M U as a symmetric ring spectrum with an action of the Barratt-Eccles operad. We shall see how to realize M U as a strictly commutative symmetric ring spectrum in Example 8.17. Example 8.9. Let as before BB denote the M-space monoid defined by the braid groups. We shall identify the map BBhN → BBhM in terms of Quillen’s plus construction. Firstly, it follows from the homological stability of the braid groups (see [11], III, Appendix) and the homological version of Lemma 4.4, that this map is a homology isomorphism. Secondly, the monoidal structure of M gives BBhM the structure of a topological monoid, which in particular implies that its fundamental group is abelian. Thus, the map in question has the effect of abelianizing the fundamental group. The space BBhN may be identified with the classifying space of the infinite braid group B(∞). Since the commutator subgroup of the latter is perfect, it follows from the above that we may identify BBhM with Quillen’s plus construction BB+ hN . It is proved in [10] that there is a homotopy commutative diagram θ BBhN −−−−→ Ω2 (S 3 ) η (Bρ) hN y y BFhN BFhN , where η denotes the “Mahowald orientation”, that is, the extension of the nontrivial map S 1 → BFhN to a 2-fold loop map. It is a theorem of Mahowald [24], that the Thom spectrum of η is stably equivalent to the Eilenberg-Mac Lane spectrum HZ/2. By the universal property of the plus construction, we conclude from the above that there is a homotopy commutative diagram ∼ BBhM −−−−→ Ω2 (S 3 ) η (Bρ) hM y y BFhI BFhI , where the upper map is a homotopy equivalence as indicated. Consequently, the symmetric ring spectrum T ((Bρ)hM ) is a model of HZ/2. Example 8.10. Let BGL(Z) be the commutative I-space monoid associated to the general linear groups GLn (Z). As in the case of the braid groups, we may identify BGL(Z)hN → BGL(Z)hI in terms of Quillen’s plus construction. Indeed, by the homological stability of the groups GLn (Z), it follows that this map is a homology equivalence. Since BGL(Z)hI is a topological monoid it has abelian fundamental group, hence it may be identified with BGL∞ (Z)+ ; the base point component of Quillen’s algebraic K-theory space. In similar fashion, starting with the I-space BΣ, we may identify BΣhI with BΣ+ ∞ , which by the Barratt-Priddy-Quillen Theorem is equivalent to the base point component of Q(S 0 ). Example 8.11. Let f : X → BF (k) be a based map and consider the associated map of I[k]-spaces X • → BF . It is proved in [34] that if X is well-based and Documenta Mathematica 14 (2009) 699–748 740 Christian Schlichtkrull • connected, then XhI is a model of the infinite loop space Q(X). Identifying • • XhI with XhI[k] via the isomorphism ψk , it follows as in Example 8.8 that • the induced map XhI[k] → BFhI is a map of E-spaces which models the usual extension of f to a map of infinite loop spaces. If we instead think of X • as an M-space by restriction, then one can show that XhM is homotopy equivalent • to the colimit XM , that is, to the free based monoid generated by X. By a • • theorem of James [18] the latter is a model of ΩΣ(X), and the map XhM → XhI corresponds to the inclusion of ΩΣ(X) in Q(X). Example 8.12. Let Ek be the kth stage of the Smith filtration of the BarrattEccles operad E and write X 7→ Ek (X) for the associated monad on based spaces, see [3], [38]. Thus, Ek is equivalent to the little k-cubes operad, and if X is a well-based connected space, then Ek (X) is a combinatorial model of Ωk Σk (X). Given a based map f : X → BFhI , we use that Ek is augmented over E to extend f to a map of Ek -spaces Ek (f ) : Ek (X) → Ek (BFhI ) → BFhI , which for connected X is a model of the usual extension of f to a k-fold loop map. It follows that the associated symmetric Thom spectrum T (Ek (f )) is equipped with an Ek -action. 8.4. Symmetrization via Kan Extension. Let again D be a monoidal subcategory of I such that D ∩ N is cofinal in N and let us write j : D → I for the inclusion. We first consider homotopy Kan extensions along j. Recall that given a D-space X, the homotopy Kan extension is the I-space j∗h (X) defined by j∗h (X)(n) = hocolim X ◦ πn , (j↓n) see Appendix A.1. The functor j∗h : j∗h (−) induces a functor DU/BF → IU/BF, (f : X → BF ) 7→ (j∗h (f ) : j∗h (X) → j∗h (BF ) → BF ) which is I-equivalent to the functor (8.5) in the sense of the following lemma. Lemma 8.13. There is a natural I-equivalence j∗h (X) → RfhD (XhD ) of Ispaces over BF . Proof. Notice first that there is a commutative diagram j∗h (X) −−−−→ y BF y XhD −−−−→ BFhI , inducing a map of I-spaces j∗h (X) → RfhD (XhD ) over BF . Since this is also a map over XhD , the result follows from the fact that j∗h (X) → XhD and RfhD (XhD ) → XhD are I-equivalences, see Lemma A.2 and the proof of Proposition 4.9. Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 741 We now turn to (categorical) Kan extensions. Given a D-space, the Kan extension j∗ (X) is defined as the homotopy Kan extension except that we use the colimit instead of the homotopy colimit, that is, j∗ (X)(n) = colim X ◦ πn . (j↓n) The functor j∗ is left adjoint to the functor that pulls an I-space back to a D-space via j and it induces a functor j∗ : DU/BF → IU/BF, (X → BF ) 7→ (j∗ (X) → j∗ (BF ) → BF ) where the map j∗ (BF ) → BF is the counit of the adjunction. This functor is strong monoidal and is symmetric monoidal if D and j are. Thus, in the latter case it takes commutative D-space monoids to commutative I-space monoids. The drawback of using the categorical Kan extension is of course that it is homotopically well-behaved only under suitable cofibration conditions on the D-space X and the main purpose of this section is to formulate such conditions. More precisely, we shall consider an inclusion j : D → I of a (not necessarily monoidal) subcategory D of I and we shall formulate conditions on D and X which ensure that the canonical map j∗h (X) → j∗ (X) is a level-wise equivalence. Given an object d0 of D, consider the category (D ↓ d0 ) of objects in D over d0 and let ∂d0 be the subcategory obtained by excluding the terminal objects. Lemma 8.14. Let j : D → I be the inclusion of a subcategory D and suppose that X is a D-space such that the map colim X ◦ πd0 → colim X ◦ πd0 = X(d0 ) ∂d0 (D↓d0 ) is a cofibration for all objects d0 in D. Then j∗h (X) → j∗ (X) is a level-wise equivalence. Proof. Notice first that the category (j ↓ n) is a preorder in the sense that the morphism sets have at most one element. Choosing a representative for each isomorphism class we get an equivalent skeleton subcategory A(n) (in fact a partially ordered set), and it suffices to show that the map hocolim X ◦ πn → colim X ◦ πn A(n) A(n) is a weak homotopy equivalence. The advantage of this is that the category A(n) is very small in the sense that its nerve only has finitely many nondegenerate simplices. In this situation there is a general model categorical criterion for comparing the homotopy colimit to the colimit, see [13], Section 10. Working in the Strøm model category on U [41], we must check that for each object a in A(n) the map colim X ◦ πn ◦ πa → colim X ◦ πn ◦ πa = X(πn (a)) ∂a (A(n)↓a) is a cofibration. Here we use the notation ∂a for the subcategory of (A(n) ↓ a) obtained by excluding the terminal object. It remains to see that if a is an object of the form d0 → n, then this criterion is the same as that stated in the Documenta Mathematica 14 (2009) 699–748 742 Christian Schlichtkrull lemma. On the one hand we may view (A(n) ↓ a) as a skeleton subcategory of ((j ↓ n) ↓ a) and on the other hand we may identify the latter category with (D ↓ d0 ). Taken together this gives a homeomorphism colim X ◦ πn ◦ πa ∼ = colim X ◦ πd0 ∂d0 ∂a and the conclusion follows. The criterion in Lemma 8.14 is not very practical and in order to have a more convenient formulation we impose conditions on the subcategory D of I. We say that D has the intersection property if each diagram in D of the form δ δ 2 1 d2 d12 ←− d1 −→ can be completed to a commutative square (8.15) d0 −−−−→ y δ d1 δ y1 d2 −−−2−→ d12 in D such that the image of the composite morphism equals the intersection of the images of δ1 and δ2 . For example, the monoidal subcategories I[k] and J [k] have the intersection property for all k ≥ 1. We say that a D-space X is intersection cofibrant if for any diagram of the form (8.15), such that the intersection of the images of δ1 and δ2 equals the image of the composite morphism, the induced map X(d1 ) ∪X(d0 ) X(d2 ) → X(d12 ) is a cofibration. By Lillig’s union theorem [21] for cofibrations, this is equivalent to the requirement that (i) any morphism d1 → d2 in D induces a cofibration X(d1 ) → X(d2 ), and (ii) that the intersection of the images of X(d1 ) and X(d2 ) in X(d12 ) equals the image of X(d0 ). Proposition 8.16. Let j : D → I be the inclusion of a subcategory D which has the intersection property and let X be a D-space which is intersection cofibrant. Then the map j∗h (X) → j∗ (X) is a level-wise equivalence. Proof. We show that the assumptions on D and X imply that the criterion in Lemma 8.14 is satisfied. Given an object d0 in D, consider the range functor r : (D ↓ d0 ) → (I ↓ d0 ) → P(d0 ), δ r(d − → d0 ) = δ(d) ⊆ d0 , where P(d0 ) denotes the category of subsets and inclusions in d0 . The assumption that D has the intersection property implies that the image of r is a full subcategory of P(d0 ) that is closed under inclusions and that r defines an equivalence of categories between (D ↓ d0 ) and its image. Thus, we might as well view X ◦ πd0 as a diagram U 7→ X(U ) indexed on the objects U in a full subcategory A of P(d0 ) that is closed under intersections. By assumption (i) Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 743 above we may view X(U ) as a closed subspace of X(d0 ) for all U ∈ A and by assumption (ii) we have the equality X(U ) ∩ X(V ) = X(U ∩ V ) for all pairs of objects U and V in A. It therefore follows from the gluing lemma for continuous functions on a union of closed subspaces that colim∂d0 X ◦ πd0 is homeomorphic to the union of the subspaces X(U ) of X(d0 ) for U 6= d0 . The conclusion then follows from Lillig’s union theorem for cofibrations [21]. Example 8.17. Let j : I[2] → I be the inclusion of the symmetric monoidal subcategory I[2]. Since I[2] has the intersection property and the commutative I[2]-space monoid BU is intersection cofibrant, it follows from Lemma 8.13 and Proposition 8.16 that there is a chain of I-equivalences ∼ ∼ → R(BUhI ) j∗ (BU ) ← − j∗h (BU ) − over BF . Thus, it follows from Proposition 8.7 together with Theorem 1.4 and Lemma 2.3 that applying the symmetric Thom spectrum functor to the commutative I-space monoid j∗ (BU ) gives a commutative symmetric ring spectrum which is a model of M U . 8.5. Orthogonal Thom spectra and diagram lifting. Recall from [27] that an orthogonal spectrum is a spectrum X such that the nth space X(n) has an action of the orthogonal group O(n), and such that the iterated structure maps S m ∧ X(n) → X(m + n) are O(m) × O(n)-equivariant. We write SpO for the category of orthogonal spectra. Let V be the topological category whose objects are the vector spaces Rn , and whose morphisms are the linear isometries. A V-space is a continuous functor V → U, and we write VU for the category of such functors. The symmetric monoidal structure of V induces a symmetric monoidal structure on VU in the usual way, and the I-space BF extends to a commutative V-space monoid. Applying the Thom space construction level-wise as in the definition of the symmetric Thom spectrum functor, we get the symmetric monoidal orthogonal Thom spectrum functor T : VU/BF → SpO . In order to construct orthogonal Thom spectra from space level data, we need a V-space version R : U/BFhV → VU/BF of the I-space lifting functor. Here the homotopy colimit BFhV denotes the realization of the simplicial space a [k] 7→ V(Rn1 , Rn0 ) × · · · × V(Rnk , Rnk−1 ) × BF (nk ). n0 ,...,nk The statement in Lemma 4.4 remains true with V instead of I, and we conclude from this that the canonical map BFhN → BFhV is a weak homotopy equivalence. The definition of the V-space lifting functor is then completely analogous to the definition of the I-space lifting functor in Section 4.2: Let BF be the V-space defined by the homotopy Kan extension along the identity functor on Documenta Mathematica 14 (2009) 699–748 744 Christian Schlichtkrull V. Given a map f : X → BFhV , we define Rf (X) to be the homotopy pullback of the diagram of V-spaces f π − BF , X− → BFhV ← and we define R(f ) to be the composition R(f ) : Rf (X) → BF → BF. The Barratt-Eccles operad acts on BFhV and the results on preservation of operad actions from Section 6 carry over to this setting. Appendix A. Homotopy colimits We here collect the facts about homotopy colimits needed in the paper. We shall adapt the definitions of Bousfield and Kan [7], except that we work with topological spaces instead of simplicial sets. Thus, given a small category A and an A-diagram X : A → U, the homotopy colimit hocolimC X is defined to be the realization of the simplicial space a (A.1) [k] 7→ X(ak ), a0 ←···←ak where the coproduct is over the k-simplices of the nerve N• C. It is sometimes convenient to view this as the classifying space of the topological category A(X) whose space of objects is the disjoint union of the spaces X(a) where a runs through the objects of A. A morphism (a, x) → (a′ , x′ ) in A(X) is specified by a morphism α : a → a′ in A such that α∗ x = x′ . If X is a based A-diagram, that is, a functor X : A → T , then the inclusion of the base points gives a map BA → BA(X) and we define the based homotopy colimit to be the quotient space. Equivalently, this is the realization of the simplicial space obtained by replacing the disjoint union in (A.1) by the wedge product. A.1. Homotopy Kan extensions. Let φ : A → B be a functor between small categories. Given an A-diagram X, the (left) homotopy Kan extension φh∗ X is the B-diagram defined by φh∗ X(b) = hocolim X ◦ πb . (φ↓b) The homotopy colimit is over the category (φ ↓ b) whose objects are pairs (a, β) in which a is an object of A and β : φ(a) → b is a morphism in B. A morphism (a, β) → (a′ , β ′ ) is given by a morphism α : a → a′ in A such that β = β ′ ◦ φ(α). The functor πb : (φ ↓ b) → A is defined by (a, β) 7→ a. We recall that the categorical Kan extension φ∗ X is defined using the categorical colimit instead of the homotopy colimit, see [22]. If B is the terminal category ∗ and p : A → ∗ the projection, then p∗ X and ph∗ X are respectively the colimit and the homotopy colimit of the A-diagram X. Notice, that the functors πb define a map of B-diagrams from φh∗ X to the constant B-diagram hocolimA X. A proof of the following well-known lemma can be found in [34]. Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 745 Lemma A.2. The induced map ∼ π : hocolim φh∗ X − → hocolim X. B A is a weak homotopy equivalence. This lemma may be viewed as a statement about the composition of two derived functors. Given an additional functor ψ : B → C, one can more generally show ∼ → (ψφ)h∗ . In the lemma that there is a natural equivalence of functors ψ∗h φh∗ − below, we shall consider the case where X has the form φ∗ Y for a B-diagram Y , and we shall relate π to the map of homotopy colimits induced by the natural transformation of B-diagrams t : φh∗ φ∗ Y → φ∗ φ∗ Y → Y, where the first arrow is the canonical projection from the homotopy colimit to the colimit and the second arrow is given by the universal property of the categorical Kan extension. Lemma A.3. Given a B-diagram Y , the diagram π / hocolim φ∗ Y hocolim φh∗ φ∗ Y A B NNN qq NNN q q NNN qqq t NN& xqqq φ hocolim Y B is homotopy commutative by a canonical choice of a natural homotopy. Proof. We may view the homotopy colimit of the B-diagram φh∗ φ∗ Y as the realization of the bisimplicial space a φ∗ Y (aj ), ([i], [j]) 7→ n b0 ←...←bi ←φ(a0 ) a0 ←...←aj o and it is well-known that this is homeomorphic to the realization of the diagonal simplicial space. Restricting to this simplicial space, the two maps in the diagram are induced by the simplicial maps that map a simplex γ α α i 1 ai , y) . . . ←− − φ(a0 ), a0 ←− (b0 ← . . . ← bi ← with y in φ∗ Y (ai ), to (b0 ← . . . ← bi , γ∗ φ(α1 . . . αi )∗ y), φ(α0 ) φ(αi ) respectively (φ(a0 ) ←−−− . . . ←−−− φ(ai ), y). The required homotopy between the topological realizations of these maps is then defined by [(b0 ← . . . ← bi ← φ(a0 ) ← . . . ← φ(ai ), y); (su, (1 − s)u)], i for u ∈ ∆ and s ∈ I. Here I denotes the unit interval and ∆i = {(u0 , . . . , ui ) ∈ I i+1 : u0 + · · · + ui = 1} Documenta Mathematica 14 (2009) 699–748 746 Christian Schlichtkrull is the standard i-simplex. The following lemma is needed to ensure that the functor R defined in Section 4.2 takes values in the subcategory of level-wise T -good objects in IU/BF . Lemma A.4. Let A be a small category and let fa : Xa → BF (n) be an Adiagram in U/BF (n). If each fa classifies a well-based quasifibration, then the induced map f : hocolim Xa → BF (n) A also classifies a well-based quasifibration. Proof. Let Wa = fa∗ V (n), and notice that f ∗ V (n) is homeomorphic to hocolimA Wa since topological realization preserves pullback diagrams. It follows that the pullback of V (n) → BF (n) along f is homeomorphic to the realization of the simplicial map a a Wak → Xak . a0 ←···←ak a0 ←···←ak These are good simplicial spaces in the sense of [36], Appendix A, and the map is a degree-wise quasifibration by assumption. The result then follows from standard results on realization of simplicial quasifibrations and simplicial cofibrations, see e.g. [36], Proposition 1.6 and [19]. References [1] R. D. Arthan and S. R. Bullett, The homology of M O(1)∧∞ and M U (1)∧∞ , J. Pure Appl. Algebra 26 (1982), no. 3, 229–234. [2] M. G. Barratt and P. J. Eccles, Γ+ -structures-I: A free group functor for stable homotopy theory, Topology 13 (1974), 25–45. [3] C. Berger, Combinatorial models for real configuration spaces and En operads, Contemp. Math., vol. 202, 37–52, Amer. Math. Soc., Providence, RI, 1997. [4] A. Blumberg, R. Cohen, and C. Schlichtkrull, Topological Hochschild homology of Thom spectra and the free loop space, Preprint, 2008. [5] J. S. Birman, Braids, links, and mapping class groups, Annals of Mathematics Studies, No 82, Princeton University Press, Princeton, N.J., 1975. [6] F. Borceux, Handbook of categorical algebra 2: Categories and structures, Encyclopedia of Mathematics and its Applications 51, Cambridge University Press, Cambridge, 1994. [7] A. K. Bousfield and D. M. Kan, Homotopy limits, completions and localizations, Springer Lecture Notes in Math. 304, Springer Verlag, 1972. [8] M. Bökstedt, Topological Hochschild homology, Preprint 1985, Bielefeld University. [9] S. R. Bullett, Permutations and braids in cobordism theory, Proc. London Math. Soc. (3) 38 (1979), no. 3, 517–531. [10] F. R. Cohen, Braid orientations and bundles with flat connection, Inventiones math. 46, (1978), 99–110. Documenta Mathematica 14 (2009) 699–748 Thom Spectra that Are Symmetric Spectra 747 [11] F. R. Cohen, T. Lada, and J. P. May, Homology of iterated loop spaces, Springer Lecture Notes in Math. 533, Springer Verlag, 1976. [12] F. R. Cohen and V. V. Vershinin, Thom spectra which are wedges of Eilenberg-Mac Lane spectra, Stable and unstable homotopy (Toronto, ON, 1996), 43–65, Fields Inst. Commun., 19, Amer. Math. Soc., Providence, RI, 1998. [13] W. G. Dwyer and J. Spalinski, Homotopy theories and model categories, Handbook of algebraic topology, North-Holland, Amsterdam, 1995, 73–126. [14] T. Goodwillie, Calculus II: Analytic functors, K-theory 5, (1992), no. 4, 295–332. [15] P. S. Hirschhorn, Model categories and their localizations, Mathematical Surveys and Monographs 99, American Mathematical Society, Providence, R.I, 2003. [16] M. Hovey, B. Shipley and J. Smith, Symmetric spectra, J. Amer. Math. Soc. 13, (2000), no. 1, 149–208. [17] J. Hollender and R. M. Vogt, Modules of topological spaces, applications to homotopy limits and E∞ structures, Arch. Math. 59 (1992), no. 2, pp. 115–129. [18] I. M. James, Reduced product spaces, Ann. of Math. (2) 62, (1955), 170– 197. [19] L. G. Lewis, When is the natural map X → ΩΣX a cofibration?, Trans. Amer. Math. Soc. 273 (1982), no. 1, 147–155. [20] L. G. Lewis, J. P. May and M. Steinberger, Equivariant stable homotopy theory, Springer Lecture Notes in Math. 1213, Springer Verlag, 1986. [21] J. Lillig, A union theorem for cofibrations, Arch. Math 24 (1973), 410–415. [22] S. Mac Lane, Categories for the working mathematician. Second Edition, Graduate Texts in Mathematics, Springer Verlag, New York, 1998. Springer-Verlag, 1971. [23] I. Madsen, Algebraic K-theory and traces, from: “Current Developments in Mathematics, 1995, (Cambridge, MA), International Press, 191–321 (1996). [24] M. Mahowald, A new infinite family in 2 π∗s , Topology 16 (1977), 249–256. [25] M. Mahowald, Ring spectra which are Thom spectra, Duke Math. J. 46 (1979), 549–559. [26] M. Mahowald and N. Ray, A note on the Thom isomorphism, Proc. Amer. Math. Soc. 82 (1981), no. 2, 307–308. [27] M. A. Mandell, J. P. May, S. Schwede and B. Shipley, Model categories of diagram spectra, Proc. London Math. Soc. (3), 82 (2001), no. 2, 441–512. [28] J. P. May, The geometry of iterated loop spaces, Springer Lecture Notes in Math. 271, Springer Verlag, 1972. [29] J. P. May, E∞ spaces, group completions, and permutative categories, New developments in topology (Proc. Sympos. Algebraic Topology, Oxford, 1972), pp. 61–93. London Math. Soc. Lecture Note Ser., No. 11, Cambridge Univ. Press, London, 1974. Documenta Mathematica 14 (2009) 699–748 748 Christian Schlichtkrull [30] J. P. May, Classifying spaces and fibrations, Mem. Amer. Math. Soc. 1 (1975), 1, no. 155. [31] J. P. May, E∞ ring spaces and E∞ ring spectra, Springer Lecture Notes in Math. 577, Springer Verlag, 1977. [32] J. P. May and J. Sigurdsson, Parametrized homotopy theory, Mathematical Surveys and Monographs, 132, American Mathematical Society, Providence, RI, 2006. [33] S. Sagave and C. Schlichtkrull, Diagram spaces and symmetric spectra, in preparation. [34] C. Schlichtkrull, The homotopy infinite symmetric product represents stable homotopy, Algebr. Geom. Topol. 7 (2007), 1963–1977. [35] S. Schwede, Symmetric spectra, in preparation, online version accessible at http://www.math.uni-bonn.de/people/schwede/SymSpec.pdf. [36] G. Segal, Categories and cohomology theories, Topology 13 (1974), 293– 312. [37] B. Shipley, Symmetric spectra and topological Hochschild homology, Ktheory 19, (2000), no. 2, 155–183. [38] J. H. Smith, Simplicial group models for Ωn S n X, Israel J. Math. 66 (1989), no. 1–3, 330–350. [39] N. Steenrod, A convenient category of topological spaces, Michigan Math. J. 14 (1967), 133–152. [40] A. Strøm, Note on cofibrations II, Math. Scand. 22 (1968), 130–142. [41] A. Strøm, The homotopy category is a homotopy category, Arch. Math. 23 (1972), 435–441. [42] F. Waldhausen, Algebraic K-theory of spaces, Lecture Notes in Math. 1126, 318–419, Springer-Verlag, New York, 1985. Christian Schlichtkrull Department of Mathematics University of Bergen Johannes Brunsgate 12 5008 Bergen Norway [email protected] Documenta Mathematica 14 (2009) 699–748